]>
Commit | Line | Data |
---|---|---|
8cdea7c0 BS |
1 | /* memcontrol.c - Memory Controller |
2 | * | |
3 | * Copyright IBM Corporation, 2007 | |
4 | * Author Balbir Singh <balbir@linux.vnet.ibm.com> | |
5 | * | |
78fb7466 PE |
6 | * Copyright 2007 OpenVZ SWsoft Inc |
7 | * Author: Pavel Emelianov <xemul@openvz.org> | |
8 | * | |
2e72b634 KS |
9 | * Memory thresholds |
10 | * Copyright (C) 2009 Nokia Corporation | |
11 | * Author: Kirill A. Shutemov | |
12 | * | |
7ae1e1d0 GC |
13 | * Kernel Memory Controller |
14 | * Copyright (C) 2012 Parallels Inc. and Google Inc. | |
15 | * Authors: Glauber Costa and Suleiman Souhlal | |
16 | * | |
1575e68b JW |
17 | * Native page reclaim |
18 | * Charge lifetime sanitation | |
19 | * Lockless page tracking & accounting | |
20 | * Unified hierarchy configuration model | |
21 | * Copyright (C) 2015 Red Hat, Inc., Johannes Weiner | |
22 | * | |
8cdea7c0 BS |
23 | * This program is free software; you can redistribute it and/or modify |
24 | * it under the terms of the GNU General Public License as published by | |
25 | * the Free Software Foundation; either version 2 of the License, or | |
26 | * (at your option) any later version. | |
27 | * | |
28 | * This program is distributed in the hope that it will be useful, | |
29 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
30 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
31 | * GNU General Public License for more details. | |
32 | */ | |
33 | ||
3e32cb2e | 34 | #include <linux/page_counter.h> |
8cdea7c0 BS |
35 | #include <linux/memcontrol.h> |
36 | #include <linux/cgroup.h> | |
78fb7466 | 37 | #include <linux/mm.h> |
6e84f315 | 38 | #include <linux/sched/mm.h> |
3a4f8a0b | 39 | #include <linux/shmem_fs.h> |
4ffef5fe | 40 | #include <linux/hugetlb.h> |
d13d1443 | 41 | #include <linux/pagemap.h> |
d52aa412 | 42 | #include <linux/smp.h> |
8a9f3ccd | 43 | #include <linux/page-flags.h> |
66e1707b | 44 | #include <linux/backing-dev.h> |
8a9f3ccd BS |
45 | #include <linux/bit_spinlock.h> |
46 | #include <linux/rcupdate.h> | |
e222432b | 47 | #include <linux/limits.h> |
b9e15baf | 48 | #include <linux/export.h> |
8c7c6e34 | 49 | #include <linux/mutex.h> |
bb4cc1a8 | 50 | #include <linux/rbtree.h> |
b6ac57d5 | 51 | #include <linux/slab.h> |
66e1707b | 52 | #include <linux/swap.h> |
02491447 | 53 | #include <linux/swapops.h> |
66e1707b | 54 | #include <linux/spinlock.h> |
2e72b634 | 55 | #include <linux/eventfd.h> |
79bd9814 | 56 | #include <linux/poll.h> |
2e72b634 | 57 | #include <linux/sort.h> |
66e1707b | 58 | #include <linux/fs.h> |
d2ceb9b7 | 59 | #include <linux/seq_file.h> |
70ddf637 | 60 | #include <linux/vmpressure.h> |
b69408e8 | 61 | #include <linux/mm_inline.h> |
5d1ea48b | 62 | #include <linux/swap_cgroup.h> |
cdec2e42 | 63 | #include <linux/cpu.h> |
158e0a2d | 64 | #include <linux/oom.h> |
0056f4e6 | 65 | #include <linux/lockdep.h> |
79bd9814 | 66 | #include <linux/file.h> |
b23afb93 | 67 | #include <linux/tracehook.h> |
08e552c6 | 68 | #include "internal.h" |
d1a4c0b3 | 69 | #include <net/sock.h> |
4bd2c1ee | 70 | #include <net/ip.h> |
f35c3a8e | 71 | #include "slab.h" |
8cdea7c0 | 72 | |
7c0f6ba6 | 73 | #include <linux/uaccess.h> |
8697d331 | 74 | |
cc8e970c KM |
75 | #include <trace/events/vmscan.h> |
76 | ||
073219e9 TH |
77 | struct cgroup_subsys memory_cgrp_subsys __read_mostly; |
78 | EXPORT_SYMBOL(memory_cgrp_subsys); | |
68ae564b | 79 | |
7d828602 JW |
80 | struct mem_cgroup *root_mem_cgroup __read_mostly; |
81 | ||
a181b0e8 | 82 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
8cdea7c0 | 83 | |
f7e1cb6e JW |
84 | /* Socket memory accounting disabled? */ |
85 | static bool cgroup_memory_nosocket; | |
86 | ||
04823c83 VD |
87 | /* Kernel memory accounting disabled? */ |
88 | static bool cgroup_memory_nokmem; | |
89 | ||
21afa38e | 90 | /* Whether the swap controller is active */ |
c255a458 | 91 | #ifdef CONFIG_MEMCG_SWAP |
c077719b | 92 | int do_swap_account __read_mostly; |
c077719b | 93 | #else |
a0db00fc | 94 | #define do_swap_account 0 |
c077719b KH |
95 | #endif |
96 | ||
7941d214 JW |
97 | /* Whether legacy memory+swap accounting is active */ |
98 | static bool do_memsw_account(void) | |
99 | { | |
100 | return !cgroup_subsys_on_dfl(memory_cgrp_subsys) && do_swap_account; | |
101 | } | |
102 | ||
71cd3113 | 103 | static const char *const mem_cgroup_lru_names[] = { |
58cf188e SZ |
104 | "inactive_anon", |
105 | "active_anon", | |
106 | "inactive_file", | |
107 | "active_file", | |
108 | "unevictable", | |
109 | }; | |
110 | ||
a0db00fc KS |
111 | #define THRESHOLDS_EVENTS_TARGET 128 |
112 | #define SOFTLIMIT_EVENTS_TARGET 1024 | |
113 | #define NUMAINFO_EVENTS_TARGET 1024 | |
e9f8974f | 114 | |
bb4cc1a8 AM |
115 | /* |
116 | * Cgroups above their limits are maintained in a RB-Tree, independent of | |
117 | * their hierarchy representation | |
118 | */ | |
119 | ||
ef8f2327 | 120 | struct mem_cgroup_tree_per_node { |
bb4cc1a8 | 121 | struct rb_root rb_root; |
fa90b2fd | 122 | struct rb_node *rb_rightmost; |
bb4cc1a8 AM |
123 | spinlock_t lock; |
124 | }; | |
125 | ||
bb4cc1a8 AM |
126 | struct mem_cgroup_tree { |
127 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; | |
128 | }; | |
129 | ||
130 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; | |
131 | ||
9490ff27 KH |
132 | /* for OOM */ |
133 | struct mem_cgroup_eventfd_list { | |
134 | struct list_head list; | |
135 | struct eventfd_ctx *eventfd; | |
136 | }; | |
2e72b634 | 137 | |
79bd9814 TH |
138 | /* |
139 | * cgroup_event represents events which userspace want to receive. | |
140 | */ | |
3bc942f3 | 141 | struct mem_cgroup_event { |
79bd9814 | 142 | /* |
59b6f873 | 143 | * memcg which the event belongs to. |
79bd9814 | 144 | */ |
59b6f873 | 145 | struct mem_cgroup *memcg; |
79bd9814 TH |
146 | /* |
147 | * eventfd to signal userspace about the event. | |
148 | */ | |
149 | struct eventfd_ctx *eventfd; | |
150 | /* | |
151 | * Each of these stored in a list by the cgroup. | |
152 | */ | |
153 | struct list_head list; | |
fba94807 TH |
154 | /* |
155 | * register_event() callback will be used to add new userspace | |
156 | * waiter for changes related to this event. Use eventfd_signal() | |
157 | * on eventfd to send notification to userspace. | |
158 | */ | |
59b6f873 | 159 | int (*register_event)(struct mem_cgroup *memcg, |
347c4a87 | 160 | struct eventfd_ctx *eventfd, const char *args); |
fba94807 TH |
161 | /* |
162 | * unregister_event() callback will be called when userspace closes | |
163 | * the eventfd or on cgroup removing. This callback must be set, | |
164 | * if you want provide notification functionality. | |
165 | */ | |
59b6f873 | 166 | void (*unregister_event)(struct mem_cgroup *memcg, |
fba94807 | 167 | struct eventfd_ctx *eventfd); |
79bd9814 TH |
168 | /* |
169 | * All fields below needed to unregister event when | |
170 | * userspace closes eventfd. | |
171 | */ | |
172 | poll_table pt; | |
173 | wait_queue_head_t *wqh; | |
ac6424b9 | 174 | wait_queue_entry_t wait; |
79bd9814 TH |
175 | struct work_struct remove; |
176 | }; | |
177 | ||
c0ff4b85 R |
178 | static void mem_cgroup_threshold(struct mem_cgroup *memcg); |
179 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg); | |
2e72b634 | 180 | |
7dc74be0 DN |
181 | /* Stuffs for move charges at task migration. */ |
182 | /* | |
1dfab5ab | 183 | * Types of charges to be moved. |
7dc74be0 | 184 | */ |
1dfab5ab JW |
185 | #define MOVE_ANON 0x1U |
186 | #define MOVE_FILE 0x2U | |
187 | #define MOVE_MASK (MOVE_ANON | MOVE_FILE) | |
7dc74be0 | 188 | |
4ffef5fe DN |
189 | /* "mc" and its members are protected by cgroup_mutex */ |
190 | static struct move_charge_struct { | |
b1dd693e | 191 | spinlock_t lock; /* for from, to */ |
264a0ae1 | 192 | struct mm_struct *mm; |
4ffef5fe DN |
193 | struct mem_cgroup *from; |
194 | struct mem_cgroup *to; | |
1dfab5ab | 195 | unsigned long flags; |
4ffef5fe | 196 | unsigned long precharge; |
854ffa8d | 197 | unsigned long moved_charge; |
483c30b5 | 198 | unsigned long moved_swap; |
8033b97c DN |
199 | struct task_struct *moving_task; /* a task moving charges */ |
200 | wait_queue_head_t waitq; /* a waitq for other context */ | |
201 | } mc = { | |
2bd9bb20 | 202 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
203 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
204 | }; | |
4ffef5fe | 205 | |
4e416953 BS |
206 | /* |
207 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
208 | * limit reclaim to prevent infinite loops, if they ever occur. | |
209 | */ | |
a0db00fc | 210 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
bb4cc1a8 | 211 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2 |
4e416953 | 212 | |
217bc319 KH |
213 | enum charge_type { |
214 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
41326c17 | 215 | MEM_CGROUP_CHARGE_TYPE_ANON, |
d13d1443 | 216 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 217 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
218 | NR_CHARGE_TYPE, |
219 | }; | |
220 | ||
8c7c6e34 | 221 | /* for encoding cft->private value on file */ |
86ae53e1 GC |
222 | enum res_type { |
223 | _MEM, | |
224 | _MEMSWAP, | |
225 | _OOM_TYPE, | |
510fc4e1 | 226 | _KMEM, |
d55f90bf | 227 | _TCP, |
86ae53e1 GC |
228 | }; |
229 | ||
a0db00fc KS |
230 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
231 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) | |
8c7c6e34 | 232 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
9490ff27 KH |
233 | /* Used for OOM nofiier */ |
234 | #define OOM_CONTROL (0) | |
8c7c6e34 | 235 | |
b05706f1 KT |
236 | /* |
237 | * Iteration constructs for visiting all cgroups (under a tree). If | |
238 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
239 | * be used for reference counting. | |
240 | */ | |
241 | #define for_each_mem_cgroup_tree(iter, root) \ | |
242 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ | |
243 | iter != NULL; \ | |
244 | iter = mem_cgroup_iter(root, iter, NULL)) | |
245 | ||
246 | #define for_each_mem_cgroup(iter) \ | |
247 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ | |
248 | iter != NULL; \ | |
249 | iter = mem_cgroup_iter(NULL, iter, NULL)) | |
250 | ||
7775face TH |
251 | static inline bool should_force_charge(void) |
252 | { | |
253 | return tsk_is_oom_victim(current) || fatal_signal_pending(current) || | |
254 | (current->flags & PF_EXITING); | |
255 | } | |
256 | ||
70ddf637 AV |
257 | /* Some nice accessors for the vmpressure. */ |
258 | struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg) | |
259 | { | |
260 | if (!memcg) | |
261 | memcg = root_mem_cgroup; | |
262 | return &memcg->vmpressure; | |
263 | } | |
264 | ||
265 | struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr) | |
266 | { | |
267 | return &container_of(vmpr, struct mem_cgroup, vmpressure)->css; | |
268 | } | |
269 | ||
84c07d11 | 270 | #ifdef CONFIG_MEMCG_KMEM |
55007d84 | 271 | /* |
f7ce3190 | 272 | * This will be the memcg's index in each cache's ->memcg_params.memcg_caches. |
b8627835 LZ |
273 | * The main reason for not using cgroup id for this: |
274 | * this works better in sparse environments, where we have a lot of memcgs, | |
275 | * but only a few kmem-limited. Or also, if we have, for instance, 200 | |
276 | * memcgs, and none but the 200th is kmem-limited, we'd have to have a | |
277 | * 200 entry array for that. | |
55007d84 | 278 | * |
dbcf73e2 VD |
279 | * The current size of the caches array is stored in memcg_nr_cache_ids. It |
280 | * will double each time we have to increase it. | |
55007d84 | 281 | */ |
dbcf73e2 VD |
282 | static DEFINE_IDA(memcg_cache_ida); |
283 | int memcg_nr_cache_ids; | |
749c5415 | 284 | |
05257a1a VD |
285 | /* Protects memcg_nr_cache_ids */ |
286 | static DECLARE_RWSEM(memcg_cache_ids_sem); | |
287 | ||
288 | void memcg_get_cache_ids(void) | |
289 | { | |
290 | down_read(&memcg_cache_ids_sem); | |
291 | } | |
292 | ||
293 | void memcg_put_cache_ids(void) | |
294 | { | |
295 | up_read(&memcg_cache_ids_sem); | |
296 | } | |
297 | ||
55007d84 GC |
298 | /* |
299 | * MIN_SIZE is different than 1, because we would like to avoid going through | |
300 | * the alloc/free process all the time. In a small machine, 4 kmem-limited | |
301 | * cgroups is a reasonable guess. In the future, it could be a parameter or | |
302 | * tunable, but that is strictly not necessary. | |
303 | * | |
b8627835 | 304 | * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get |
55007d84 GC |
305 | * this constant directly from cgroup, but it is understandable that this is |
306 | * better kept as an internal representation in cgroup.c. In any case, the | |
b8627835 | 307 | * cgrp_id space is not getting any smaller, and we don't have to necessarily |
55007d84 GC |
308 | * increase ours as well if it increases. |
309 | */ | |
310 | #define MEMCG_CACHES_MIN_SIZE 4 | |
b8627835 | 311 | #define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX |
55007d84 | 312 | |
d7f25f8a GC |
313 | /* |
314 | * A lot of the calls to the cache allocation functions are expected to be | |
315 | * inlined by the compiler. Since the calls to memcg_kmem_get_cache are | |
316 | * conditional to this static branch, we'll have to allow modules that does | |
317 | * kmem_cache_alloc and the such to see this symbol as well | |
318 | */ | |
ef12947c | 319 | DEFINE_STATIC_KEY_FALSE(memcg_kmem_enabled_key); |
d7f25f8a | 320 | EXPORT_SYMBOL(memcg_kmem_enabled_key); |
a8964b9b | 321 | |
17cc4dfe TH |
322 | struct workqueue_struct *memcg_kmem_cache_wq; |
323 | ||
0a4465d3 KT |
324 | static int memcg_shrinker_map_size; |
325 | static DEFINE_MUTEX(memcg_shrinker_map_mutex); | |
326 | ||
327 | static void memcg_free_shrinker_map_rcu(struct rcu_head *head) | |
328 | { | |
329 | kvfree(container_of(head, struct memcg_shrinker_map, rcu)); | |
330 | } | |
331 | ||
332 | static int memcg_expand_one_shrinker_map(struct mem_cgroup *memcg, | |
333 | int size, int old_size) | |
334 | { | |
335 | struct memcg_shrinker_map *new, *old; | |
336 | int nid; | |
337 | ||
338 | lockdep_assert_held(&memcg_shrinker_map_mutex); | |
339 | ||
340 | for_each_node(nid) { | |
341 | old = rcu_dereference_protected( | |
342 | mem_cgroup_nodeinfo(memcg, nid)->shrinker_map, true); | |
343 | /* Not yet online memcg */ | |
344 | if (!old) | |
345 | return 0; | |
346 | ||
347 | new = kvmalloc(sizeof(*new) + size, GFP_KERNEL); | |
348 | if (!new) | |
349 | return -ENOMEM; | |
350 | ||
351 | /* Set all old bits, clear all new bits */ | |
352 | memset(new->map, (int)0xff, old_size); | |
353 | memset((void *)new->map + old_size, 0, size - old_size); | |
354 | ||
355 | rcu_assign_pointer(memcg->nodeinfo[nid]->shrinker_map, new); | |
356 | call_rcu(&old->rcu, memcg_free_shrinker_map_rcu); | |
357 | } | |
358 | ||
359 | return 0; | |
360 | } | |
361 | ||
362 | static void memcg_free_shrinker_maps(struct mem_cgroup *memcg) | |
363 | { | |
364 | struct mem_cgroup_per_node *pn; | |
365 | struct memcg_shrinker_map *map; | |
366 | int nid; | |
367 | ||
368 | if (mem_cgroup_is_root(memcg)) | |
369 | return; | |
370 | ||
371 | for_each_node(nid) { | |
372 | pn = mem_cgroup_nodeinfo(memcg, nid); | |
373 | map = rcu_dereference_protected(pn->shrinker_map, true); | |
374 | if (map) | |
375 | kvfree(map); | |
376 | rcu_assign_pointer(pn->shrinker_map, NULL); | |
377 | } | |
378 | } | |
379 | ||
380 | static int memcg_alloc_shrinker_maps(struct mem_cgroup *memcg) | |
381 | { | |
382 | struct memcg_shrinker_map *map; | |
383 | int nid, size, ret = 0; | |
384 | ||
385 | if (mem_cgroup_is_root(memcg)) | |
386 | return 0; | |
387 | ||
388 | mutex_lock(&memcg_shrinker_map_mutex); | |
389 | size = memcg_shrinker_map_size; | |
390 | for_each_node(nid) { | |
391 | map = kvzalloc(sizeof(*map) + size, GFP_KERNEL); | |
392 | if (!map) { | |
393 | memcg_free_shrinker_maps(memcg); | |
394 | ret = -ENOMEM; | |
395 | break; | |
396 | } | |
397 | rcu_assign_pointer(memcg->nodeinfo[nid]->shrinker_map, map); | |
398 | } | |
399 | mutex_unlock(&memcg_shrinker_map_mutex); | |
400 | ||
401 | return ret; | |
402 | } | |
403 | ||
404 | int memcg_expand_shrinker_maps(int new_id) | |
405 | { | |
406 | int size, old_size, ret = 0; | |
407 | struct mem_cgroup *memcg; | |
408 | ||
409 | size = DIV_ROUND_UP(new_id + 1, BITS_PER_LONG) * sizeof(unsigned long); | |
410 | old_size = memcg_shrinker_map_size; | |
411 | if (size <= old_size) | |
412 | return 0; | |
413 | ||
414 | mutex_lock(&memcg_shrinker_map_mutex); | |
415 | if (!root_mem_cgroup) | |
416 | goto unlock; | |
417 | ||
418 | for_each_mem_cgroup(memcg) { | |
419 | if (mem_cgroup_is_root(memcg)) | |
420 | continue; | |
421 | ret = memcg_expand_one_shrinker_map(memcg, size, old_size); | |
422 | if (ret) | |
423 | goto unlock; | |
424 | } | |
425 | unlock: | |
426 | if (!ret) | |
427 | memcg_shrinker_map_size = size; | |
428 | mutex_unlock(&memcg_shrinker_map_mutex); | |
429 | return ret; | |
430 | } | |
fae91d6d KT |
431 | |
432 | void memcg_set_shrinker_bit(struct mem_cgroup *memcg, int nid, int shrinker_id) | |
433 | { | |
434 | if (shrinker_id >= 0 && memcg && !mem_cgroup_is_root(memcg)) { | |
435 | struct memcg_shrinker_map *map; | |
436 | ||
437 | rcu_read_lock(); | |
438 | map = rcu_dereference(memcg->nodeinfo[nid]->shrinker_map); | |
f90280d6 KT |
439 | /* Pairs with smp mb in shrink_slab() */ |
440 | smp_mb__before_atomic(); | |
fae91d6d KT |
441 | set_bit(shrinker_id, map->map); |
442 | rcu_read_unlock(); | |
443 | } | |
444 | } | |
445 | ||
0a4465d3 KT |
446 | #else /* CONFIG_MEMCG_KMEM */ |
447 | static int memcg_alloc_shrinker_maps(struct mem_cgroup *memcg) | |
448 | { | |
449 | return 0; | |
450 | } | |
451 | static void memcg_free_shrinker_maps(struct mem_cgroup *memcg) { } | |
84c07d11 | 452 | #endif /* CONFIG_MEMCG_KMEM */ |
a8964b9b | 453 | |
ad7fa852 TH |
454 | /** |
455 | * mem_cgroup_css_from_page - css of the memcg associated with a page | |
456 | * @page: page of interest | |
457 | * | |
458 | * If memcg is bound to the default hierarchy, css of the memcg associated | |
459 | * with @page is returned. The returned css remains associated with @page | |
460 | * until it is released. | |
461 | * | |
462 | * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup | |
463 | * is returned. | |
ad7fa852 TH |
464 | */ |
465 | struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page) | |
466 | { | |
467 | struct mem_cgroup *memcg; | |
468 | ||
ad7fa852 TH |
469 | memcg = page->mem_cgroup; |
470 | ||
9e10a130 | 471 | if (!memcg || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
ad7fa852 TH |
472 | memcg = root_mem_cgroup; |
473 | ||
ad7fa852 TH |
474 | return &memcg->css; |
475 | } | |
476 | ||
2fc04524 VD |
477 | /** |
478 | * page_cgroup_ino - return inode number of the memcg a page is charged to | |
479 | * @page: the page | |
480 | * | |
481 | * Look up the closest online ancestor of the memory cgroup @page is charged to | |
482 | * and return its inode number or 0 if @page is not charged to any cgroup. It | |
483 | * is safe to call this function without holding a reference to @page. | |
484 | * | |
485 | * Note, this function is inherently racy, because there is nothing to prevent | |
486 | * the cgroup inode from getting torn down and potentially reallocated a moment | |
487 | * after page_cgroup_ino() returns, so it only should be used by callers that | |
488 | * do not care (such as procfs interfaces). | |
489 | */ | |
490 | ino_t page_cgroup_ino(struct page *page) | |
491 | { | |
492 | struct mem_cgroup *memcg; | |
493 | unsigned long ino = 0; | |
494 | ||
495 | rcu_read_lock(); | |
496 | memcg = READ_ONCE(page->mem_cgroup); | |
497 | while (memcg && !(memcg->css.flags & CSS_ONLINE)) | |
498 | memcg = parent_mem_cgroup(memcg); | |
499 | if (memcg) | |
500 | ino = cgroup_ino(memcg->css.cgroup); | |
501 | rcu_read_unlock(); | |
502 | return ino; | |
503 | } | |
504 | ||
ef8f2327 MG |
505 | static struct mem_cgroup_per_node * |
506 | mem_cgroup_page_nodeinfo(struct mem_cgroup *memcg, struct page *page) | |
f64c3f54 | 507 | { |
97a6c37b | 508 | int nid = page_to_nid(page); |
f64c3f54 | 509 | |
ef8f2327 | 510 | return memcg->nodeinfo[nid]; |
f64c3f54 BS |
511 | } |
512 | ||
ef8f2327 MG |
513 | static struct mem_cgroup_tree_per_node * |
514 | soft_limit_tree_node(int nid) | |
bb4cc1a8 | 515 | { |
ef8f2327 | 516 | return soft_limit_tree.rb_tree_per_node[nid]; |
bb4cc1a8 AM |
517 | } |
518 | ||
ef8f2327 | 519 | static struct mem_cgroup_tree_per_node * |
bb4cc1a8 AM |
520 | soft_limit_tree_from_page(struct page *page) |
521 | { | |
522 | int nid = page_to_nid(page); | |
bb4cc1a8 | 523 | |
ef8f2327 | 524 | return soft_limit_tree.rb_tree_per_node[nid]; |
bb4cc1a8 AM |
525 | } |
526 | ||
ef8f2327 MG |
527 | static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_node *mz, |
528 | struct mem_cgroup_tree_per_node *mctz, | |
3e32cb2e | 529 | unsigned long new_usage_in_excess) |
bb4cc1a8 AM |
530 | { |
531 | struct rb_node **p = &mctz->rb_root.rb_node; | |
532 | struct rb_node *parent = NULL; | |
ef8f2327 | 533 | struct mem_cgroup_per_node *mz_node; |
fa90b2fd | 534 | bool rightmost = true; |
bb4cc1a8 AM |
535 | |
536 | if (mz->on_tree) | |
537 | return; | |
538 | ||
539 | mz->usage_in_excess = new_usage_in_excess; | |
540 | if (!mz->usage_in_excess) | |
541 | return; | |
542 | while (*p) { | |
543 | parent = *p; | |
ef8f2327 | 544 | mz_node = rb_entry(parent, struct mem_cgroup_per_node, |
bb4cc1a8 | 545 | tree_node); |
fa90b2fd | 546 | if (mz->usage_in_excess < mz_node->usage_in_excess) { |
bb4cc1a8 | 547 | p = &(*p)->rb_left; |
fa90b2fd DB |
548 | rightmost = false; |
549 | } | |
550 | ||
bb4cc1a8 AM |
551 | /* |
552 | * We can't avoid mem cgroups that are over their soft | |
553 | * limit by the same amount | |
554 | */ | |
555 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
556 | p = &(*p)->rb_right; | |
557 | } | |
fa90b2fd DB |
558 | |
559 | if (rightmost) | |
560 | mctz->rb_rightmost = &mz->tree_node; | |
561 | ||
bb4cc1a8 AM |
562 | rb_link_node(&mz->tree_node, parent, p); |
563 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
564 | mz->on_tree = true; | |
565 | } | |
566 | ||
ef8f2327 MG |
567 | static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
568 | struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 AM |
569 | { |
570 | if (!mz->on_tree) | |
571 | return; | |
fa90b2fd DB |
572 | |
573 | if (&mz->tree_node == mctz->rb_rightmost) | |
574 | mctz->rb_rightmost = rb_prev(&mz->tree_node); | |
575 | ||
bb4cc1a8 AM |
576 | rb_erase(&mz->tree_node, &mctz->rb_root); |
577 | mz->on_tree = false; | |
578 | } | |
579 | ||
ef8f2327 MG |
580 | static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
581 | struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 582 | { |
0a31bc97 JW |
583 | unsigned long flags; |
584 | ||
585 | spin_lock_irqsave(&mctz->lock, flags); | |
cf2c8127 | 586 | __mem_cgroup_remove_exceeded(mz, mctz); |
0a31bc97 | 587 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
588 | } |
589 | ||
3e32cb2e JW |
590 | static unsigned long soft_limit_excess(struct mem_cgroup *memcg) |
591 | { | |
592 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
4db0c3c2 | 593 | unsigned long soft_limit = READ_ONCE(memcg->soft_limit); |
3e32cb2e JW |
594 | unsigned long excess = 0; |
595 | ||
596 | if (nr_pages > soft_limit) | |
597 | excess = nr_pages - soft_limit; | |
598 | ||
599 | return excess; | |
600 | } | |
bb4cc1a8 AM |
601 | |
602 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) | |
603 | { | |
3e32cb2e | 604 | unsigned long excess; |
ef8f2327 MG |
605 | struct mem_cgroup_per_node *mz; |
606 | struct mem_cgroup_tree_per_node *mctz; | |
bb4cc1a8 | 607 | |
e231875b | 608 | mctz = soft_limit_tree_from_page(page); |
bfc7228b LD |
609 | if (!mctz) |
610 | return; | |
bb4cc1a8 AM |
611 | /* |
612 | * Necessary to update all ancestors when hierarchy is used. | |
613 | * because their event counter is not touched. | |
614 | */ | |
615 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
ef8f2327 | 616 | mz = mem_cgroup_page_nodeinfo(memcg, page); |
3e32cb2e | 617 | excess = soft_limit_excess(memcg); |
bb4cc1a8 AM |
618 | /* |
619 | * We have to update the tree if mz is on RB-tree or | |
620 | * mem is over its softlimit. | |
621 | */ | |
622 | if (excess || mz->on_tree) { | |
0a31bc97 JW |
623 | unsigned long flags; |
624 | ||
625 | spin_lock_irqsave(&mctz->lock, flags); | |
bb4cc1a8 AM |
626 | /* if on-tree, remove it */ |
627 | if (mz->on_tree) | |
cf2c8127 | 628 | __mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
629 | /* |
630 | * Insert again. mz->usage_in_excess will be updated. | |
631 | * If excess is 0, no tree ops. | |
632 | */ | |
cf2c8127 | 633 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 634 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
635 | } |
636 | } | |
637 | } | |
638 | ||
639 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) | |
640 | { | |
ef8f2327 MG |
641 | struct mem_cgroup_tree_per_node *mctz; |
642 | struct mem_cgroup_per_node *mz; | |
643 | int nid; | |
bb4cc1a8 | 644 | |
e231875b | 645 | for_each_node(nid) { |
ef8f2327 MG |
646 | mz = mem_cgroup_nodeinfo(memcg, nid); |
647 | mctz = soft_limit_tree_node(nid); | |
bfc7228b LD |
648 | if (mctz) |
649 | mem_cgroup_remove_exceeded(mz, mctz); | |
bb4cc1a8 AM |
650 | } |
651 | } | |
652 | ||
ef8f2327 MG |
653 | static struct mem_cgroup_per_node * |
654 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 655 | { |
ef8f2327 | 656 | struct mem_cgroup_per_node *mz; |
bb4cc1a8 AM |
657 | |
658 | retry: | |
659 | mz = NULL; | |
fa90b2fd | 660 | if (!mctz->rb_rightmost) |
bb4cc1a8 AM |
661 | goto done; /* Nothing to reclaim from */ |
662 | ||
fa90b2fd DB |
663 | mz = rb_entry(mctz->rb_rightmost, |
664 | struct mem_cgroup_per_node, tree_node); | |
bb4cc1a8 AM |
665 | /* |
666 | * Remove the node now but someone else can add it back, | |
667 | * we will to add it back at the end of reclaim to its correct | |
668 | * position in the tree. | |
669 | */ | |
cf2c8127 | 670 | __mem_cgroup_remove_exceeded(mz, mctz); |
3e32cb2e | 671 | if (!soft_limit_excess(mz->memcg) || |
ec903c0c | 672 | !css_tryget_online(&mz->memcg->css)) |
bb4cc1a8 AM |
673 | goto retry; |
674 | done: | |
675 | return mz; | |
676 | } | |
677 | ||
ef8f2327 MG |
678 | static struct mem_cgroup_per_node * |
679 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 680 | { |
ef8f2327 | 681 | struct mem_cgroup_per_node *mz; |
bb4cc1a8 | 682 | |
0a31bc97 | 683 | spin_lock_irq(&mctz->lock); |
bb4cc1a8 | 684 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
0a31bc97 | 685 | spin_unlock_irq(&mctz->lock); |
bb4cc1a8 AM |
686 | return mz; |
687 | } | |
688 | ||
ccda7f43 | 689 | static unsigned long memcg_sum_events(struct mem_cgroup *memcg, |
04fecbf5 | 690 | int event) |
e9f8974f | 691 | { |
a983b5eb | 692 | return atomic_long_read(&memcg->events[event]); |
e9f8974f JW |
693 | } |
694 | ||
c0ff4b85 | 695 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b070e65c | 696 | struct page *page, |
f627c2f5 | 697 | bool compound, int nr_pages) |
d52aa412 | 698 | { |
b2402857 KH |
699 | /* |
700 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is | |
701 | * counted as CACHE even if it's on ANON LRU. | |
702 | */ | |
0a31bc97 | 703 | if (PageAnon(page)) |
c9019e9b | 704 | __mod_memcg_state(memcg, MEMCG_RSS, nr_pages); |
9a4caf1e | 705 | else { |
c9019e9b | 706 | __mod_memcg_state(memcg, MEMCG_CACHE, nr_pages); |
9a4caf1e | 707 | if (PageSwapBacked(page)) |
c9019e9b | 708 | __mod_memcg_state(memcg, NR_SHMEM, nr_pages); |
9a4caf1e | 709 | } |
55e462b0 | 710 | |
f627c2f5 KS |
711 | if (compound) { |
712 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
c9019e9b | 713 | __mod_memcg_state(memcg, MEMCG_RSS_HUGE, nr_pages); |
f627c2f5 | 714 | } |
b070e65c | 715 | |
e401f176 KH |
716 | /* pagein of a big page is an event. So, ignore page size */ |
717 | if (nr_pages > 0) | |
c9019e9b | 718 | __count_memcg_events(memcg, PGPGIN, 1); |
3751d604 | 719 | else { |
c9019e9b | 720 | __count_memcg_events(memcg, PGPGOUT, 1); |
3751d604 KH |
721 | nr_pages = -nr_pages; /* for event */ |
722 | } | |
e401f176 | 723 | |
a983b5eb | 724 | __this_cpu_add(memcg->stat_cpu->nr_page_events, nr_pages); |
6d12e2d8 KH |
725 | } |
726 | ||
0a6b76dd VD |
727 | unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, |
728 | int nid, unsigned int lru_mask) | |
bb2a0de9 | 729 | { |
b4536f0c | 730 | struct lruvec *lruvec = mem_cgroup_lruvec(NODE_DATA(nid), memcg); |
e231875b | 731 | unsigned long nr = 0; |
ef8f2327 | 732 | enum lru_list lru; |
889976db | 733 | |
e231875b | 734 | VM_BUG_ON((unsigned)nid >= nr_node_ids); |
bb2a0de9 | 735 | |
ef8f2327 MG |
736 | for_each_lru(lru) { |
737 | if (!(BIT(lru) & lru_mask)) | |
738 | continue; | |
b4536f0c | 739 | nr += mem_cgroup_get_lru_size(lruvec, lru); |
e231875b JZ |
740 | } |
741 | return nr; | |
889976db | 742 | } |
bb2a0de9 | 743 | |
c0ff4b85 | 744 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 | 745 | unsigned int lru_mask) |
6d12e2d8 | 746 | { |
e231875b | 747 | unsigned long nr = 0; |
889976db | 748 | int nid; |
6d12e2d8 | 749 | |
31aaea4a | 750 | for_each_node_state(nid, N_MEMORY) |
e231875b JZ |
751 | nr += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask); |
752 | return nr; | |
d52aa412 KH |
753 | } |
754 | ||
f53d7ce3 JW |
755 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
756 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
757 | { |
758 | unsigned long val, next; | |
759 | ||
a983b5eb JW |
760 | val = __this_cpu_read(memcg->stat_cpu->nr_page_events); |
761 | next = __this_cpu_read(memcg->stat_cpu->targets[target]); | |
7a159cc9 | 762 | /* from time_after() in jiffies.h */ |
6a1a8b80 | 763 | if ((long)(next - val) < 0) { |
f53d7ce3 JW |
764 | switch (target) { |
765 | case MEM_CGROUP_TARGET_THRESH: | |
766 | next = val + THRESHOLDS_EVENTS_TARGET; | |
767 | break; | |
bb4cc1a8 AM |
768 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
769 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
770 | break; | |
f53d7ce3 JW |
771 | case MEM_CGROUP_TARGET_NUMAINFO: |
772 | next = val + NUMAINFO_EVENTS_TARGET; | |
773 | break; | |
774 | default: | |
775 | break; | |
776 | } | |
a983b5eb | 777 | __this_cpu_write(memcg->stat_cpu->targets[target], next); |
f53d7ce3 | 778 | return true; |
7a159cc9 | 779 | } |
f53d7ce3 | 780 | return false; |
d2265e6f KH |
781 | } |
782 | ||
783 | /* | |
784 | * Check events in order. | |
785 | * | |
786 | */ | |
c0ff4b85 | 787 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f KH |
788 | { |
789 | /* threshold event is triggered in finer grain than soft limit */ | |
f53d7ce3 JW |
790 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
791 | MEM_CGROUP_TARGET_THRESH))) { | |
bb4cc1a8 | 792 | bool do_softlimit; |
82b3f2a7 | 793 | bool do_numainfo __maybe_unused; |
f53d7ce3 | 794 | |
bb4cc1a8 AM |
795 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
796 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
f53d7ce3 JW |
797 | #if MAX_NUMNODES > 1 |
798 | do_numainfo = mem_cgroup_event_ratelimit(memcg, | |
799 | MEM_CGROUP_TARGET_NUMAINFO); | |
800 | #endif | |
c0ff4b85 | 801 | mem_cgroup_threshold(memcg); |
bb4cc1a8 AM |
802 | if (unlikely(do_softlimit)) |
803 | mem_cgroup_update_tree(memcg, page); | |
453a9bf3 | 804 | #if MAX_NUMNODES > 1 |
f53d7ce3 | 805 | if (unlikely(do_numainfo)) |
c0ff4b85 | 806 | atomic_inc(&memcg->numainfo_events); |
453a9bf3 | 807 | #endif |
0a31bc97 | 808 | } |
d2265e6f KH |
809 | } |
810 | ||
cf475ad2 | 811 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 812 | { |
31a78f23 BS |
813 | /* |
814 | * mm_update_next_owner() may clear mm->owner to NULL | |
815 | * if it races with swapoff, page migration, etc. | |
816 | * So this can be called with p == NULL. | |
817 | */ | |
818 | if (unlikely(!p)) | |
819 | return NULL; | |
820 | ||
073219e9 | 821 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
78fb7466 | 822 | } |
33398cf2 | 823 | EXPORT_SYMBOL(mem_cgroup_from_task); |
78fb7466 | 824 | |
d46eb14b SB |
825 | /** |
826 | * get_mem_cgroup_from_mm: Obtain a reference on given mm_struct's memcg. | |
827 | * @mm: mm from which memcg should be extracted. It can be NULL. | |
828 | * | |
829 | * Obtain a reference on mm->memcg and returns it if successful. Otherwise | |
830 | * root_mem_cgroup is returned. However if mem_cgroup is disabled, NULL is | |
831 | * returned. | |
832 | */ | |
833 | struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) | |
54595fe2 | 834 | { |
d46eb14b SB |
835 | struct mem_cgroup *memcg; |
836 | ||
837 | if (mem_cgroup_disabled()) | |
838 | return NULL; | |
0b7f569e | 839 | |
54595fe2 KH |
840 | rcu_read_lock(); |
841 | do { | |
6f6acb00 MH |
842 | /* |
843 | * Page cache insertions can happen withou an | |
844 | * actual mm context, e.g. during disk probing | |
845 | * on boot, loopback IO, acct() writes etc. | |
846 | */ | |
847 | if (unlikely(!mm)) | |
df381975 | 848 | memcg = root_mem_cgroup; |
6f6acb00 MH |
849 | else { |
850 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
851 | if (unlikely(!memcg)) | |
852 | memcg = root_mem_cgroup; | |
853 | } | |
ec903c0c | 854 | } while (!css_tryget_online(&memcg->css)); |
54595fe2 | 855 | rcu_read_unlock(); |
c0ff4b85 | 856 | return memcg; |
54595fe2 | 857 | } |
d46eb14b SB |
858 | EXPORT_SYMBOL(get_mem_cgroup_from_mm); |
859 | ||
f745c6f5 SB |
860 | /** |
861 | * get_mem_cgroup_from_page: Obtain a reference on given page's memcg. | |
862 | * @page: page from which memcg should be extracted. | |
863 | * | |
864 | * Obtain a reference on page->memcg and returns it if successful. Otherwise | |
865 | * root_mem_cgroup is returned. | |
866 | */ | |
867 | struct mem_cgroup *get_mem_cgroup_from_page(struct page *page) | |
868 | { | |
869 | struct mem_cgroup *memcg = page->mem_cgroup; | |
870 | ||
871 | if (mem_cgroup_disabled()) | |
872 | return NULL; | |
873 | ||
874 | rcu_read_lock(); | |
875 | if (!memcg || !css_tryget_online(&memcg->css)) | |
876 | memcg = root_mem_cgroup; | |
877 | rcu_read_unlock(); | |
878 | return memcg; | |
879 | } | |
880 | EXPORT_SYMBOL(get_mem_cgroup_from_page); | |
881 | ||
d46eb14b SB |
882 | /** |
883 | * If current->active_memcg is non-NULL, do not fallback to current->mm->memcg. | |
884 | */ | |
885 | static __always_inline struct mem_cgroup *get_mem_cgroup_from_current(void) | |
886 | { | |
887 | if (unlikely(current->active_memcg)) { | |
888 | struct mem_cgroup *memcg = root_mem_cgroup; | |
889 | ||
890 | rcu_read_lock(); | |
891 | if (css_tryget_online(¤t->active_memcg->css)) | |
892 | memcg = current->active_memcg; | |
893 | rcu_read_unlock(); | |
894 | return memcg; | |
895 | } | |
896 | return get_mem_cgroup_from_mm(current->mm); | |
897 | } | |
54595fe2 | 898 | |
5660048c JW |
899 | /** |
900 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
901 | * @root: hierarchy root | |
902 | * @prev: previously returned memcg, NULL on first invocation | |
903 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
904 | * | |
905 | * Returns references to children of the hierarchy below @root, or | |
906 | * @root itself, or %NULL after a full round-trip. | |
907 | * | |
908 | * Caller must pass the return value in @prev on subsequent | |
909 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
910 | * to cancel a hierarchy walk before the round-trip is complete. | |
911 | * | |
b213b54f | 912 | * Reclaimers can specify a node and a priority level in @reclaim to |
5660048c | 913 | * divide up the memcgs in the hierarchy among all concurrent |
b213b54f | 914 | * reclaimers operating on the same node and priority. |
5660048c | 915 | */ |
694fbc0f | 916 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
5660048c | 917 | struct mem_cgroup *prev, |
694fbc0f | 918 | struct mem_cgroup_reclaim_cookie *reclaim) |
14067bb3 | 919 | { |
33398cf2 | 920 | struct mem_cgroup_reclaim_iter *uninitialized_var(iter); |
5ac8fb31 | 921 | struct cgroup_subsys_state *css = NULL; |
9f3a0d09 | 922 | struct mem_cgroup *memcg = NULL; |
5ac8fb31 | 923 | struct mem_cgroup *pos = NULL; |
711d3d2c | 924 | |
694fbc0f AM |
925 | if (mem_cgroup_disabled()) |
926 | return NULL; | |
5660048c | 927 | |
9f3a0d09 JW |
928 | if (!root) |
929 | root = root_mem_cgroup; | |
7d74b06f | 930 | |
9f3a0d09 | 931 | if (prev && !reclaim) |
5ac8fb31 | 932 | pos = prev; |
14067bb3 | 933 | |
9f3a0d09 JW |
934 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
935 | if (prev) | |
5ac8fb31 | 936 | goto out; |
694fbc0f | 937 | return root; |
9f3a0d09 | 938 | } |
14067bb3 | 939 | |
542f85f9 | 940 | rcu_read_lock(); |
5f578161 | 941 | |
5ac8fb31 | 942 | if (reclaim) { |
ef8f2327 | 943 | struct mem_cgroup_per_node *mz; |
5ac8fb31 | 944 | |
ef8f2327 | 945 | mz = mem_cgroup_nodeinfo(root, reclaim->pgdat->node_id); |
5ac8fb31 JW |
946 | iter = &mz->iter[reclaim->priority]; |
947 | ||
948 | if (prev && reclaim->generation != iter->generation) | |
949 | goto out_unlock; | |
950 | ||
6df38689 | 951 | while (1) { |
4db0c3c2 | 952 | pos = READ_ONCE(iter->position); |
6df38689 VD |
953 | if (!pos || css_tryget(&pos->css)) |
954 | break; | |
5ac8fb31 | 955 | /* |
6df38689 VD |
956 | * css reference reached zero, so iter->position will |
957 | * be cleared by ->css_released. However, we should not | |
958 | * rely on this happening soon, because ->css_released | |
959 | * is called from a work queue, and by busy-waiting we | |
960 | * might block it. So we clear iter->position right | |
961 | * away. | |
5ac8fb31 | 962 | */ |
6df38689 VD |
963 | (void)cmpxchg(&iter->position, pos, NULL); |
964 | } | |
5ac8fb31 JW |
965 | } |
966 | ||
967 | if (pos) | |
968 | css = &pos->css; | |
969 | ||
970 | for (;;) { | |
971 | css = css_next_descendant_pre(css, &root->css); | |
972 | if (!css) { | |
973 | /* | |
974 | * Reclaimers share the hierarchy walk, and a | |
975 | * new one might jump in right at the end of | |
976 | * the hierarchy - make sure they see at least | |
977 | * one group and restart from the beginning. | |
978 | */ | |
979 | if (!prev) | |
980 | continue; | |
981 | break; | |
527a5ec9 | 982 | } |
7d74b06f | 983 | |
5ac8fb31 JW |
984 | /* |
985 | * Verify the css and acquire a reference. The root | |
986 | * is provided by the caller, so we know it's alive | |
987 | * and kicking, and don't take an extra reference. | |
988 | */ | |
989 | memcg = mem_cgroup_from_css(css); | |
14067bb3 | 990 | |
5ac8fb31 JW |
991 | if (css == &root->css) |
992 | break; | |
14067bb3 | 993 | |
0b8f73e1 JW |
994 | if (css_tryget(css)) |
995 | break; | |
9f3a0d09 | 996 | |
5ac8fb31 | 997 | memcg = NULL; |
9f3a0d09 | 998 | } |
5ac8fb31 JW |
999 | |
1000 | if (reclaim) { | |
5ac8fb31 | 1001 | /* |
6df38689 VD |
1002 | * The position could have already been updated by a competing |
1003 | * thread, so check that the value hasn't changed since we read | |
1004 | * it to avoid reclaiming from the same cgroup twice. | |
5ac8fb31 | 1005 | */ |
6df38689 VD |
1006 | (void)cmpxchg(&iter->position, pos, memcg); |
1007 | ||
5ac8fb31 JW |
1008 | if (pos) |
1009 | css_put(&pos->css); | |
1010 | ||
1011 | if (!memcg) | |
1012 | iter->generation++; | |
1013 | else if (!prev) | |
1014 | reclaim->generation = iter->generation; | |
9f3a0d09 | 1015 | } |
5ac8fb31 | 1016 | |
542f85f9 MH |
1017 | out_unlock: |
1018 | rcu_read_unlock(); | |
5ac8fb31 | 1019 | out: |
c40046f3 MH |
1020 | if (prev && prev != root) |
1021 | css_put(&prev->css); | |
1022 | ||
9f3a0d09 | 1023 | return memcg; |
14067bb3 | 1024 | } |
7d74b06f | 1025 | |
5660048c JW |
1026 | /** |
1027 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
1028 | * @root: hierarchy root | |
1029 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
1030 | */ | |
1031 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
1032 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
1033 | { |
1034 | if (!root) | |
1035 | root = root_mem_cgroup; | |
1036 | if (prev && prev != root) | |
1037 | css_put(&prev->css); | |
1038 | } | |
7d74b06f | 1039 | |
6df38689 VD |
1040 | static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg) |
1041 | { | |
1042 | struct mem_cgroup *memcg = dead_memcg; | |
1043 | struct mem_cgroup_reclaim_iter *iter; | |
ef8f2327 MG |
1044 | struct mem_cgroup_per_node *mz; |
1045 | int nid; | |
6df38689 VD |
1046 | int i; |
1047 | ||
9f15bde6 | 1048 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { |
6df38689 | 1049 | for_each_node(nid) { |
ef8f2327 MG |
1050 | mz = mem_cgroup_nodeinfo(memcg, nid); |
1051 | for (i = 0; i <= DEF_PRIORITY; i++) { | |
1052 | iter = &mz->iter[i]; | |
1053 | cmpxchg(&iter->position, | |
1054 | dead_memcg, NULL); | |
6df38689 VD |
1055 | } |
1056 | } | |
1057 | } | |
1058 | } | |
1059 | ||
7c5f64f8 VD |
1060 | /** |
1061 | * mem_cgroup_scan_tasks - iterate over tasks of a memory cgroup hierarchy | |
1062 | * @memcg: hierarchy root | |
1063 | * @fn: function to call for each task | |
1064 | * @arg: argument passed to @fn | |
1065 | * | |
1066 | * This function iterates over tasks attached to @memcg or to any of its | |
1067 | * descendants and calls @fn for each task. If @fn returns a non-zero | |
1068 | * value, the function breaks the iteration loop and returns the value. | |
1069 | * Otherwise, it will iterate over all tasks and return 0. | |
1070 | * | |
1071 | * This function must not be called for the root memory cgroup. | |
1072 | */ | |
1073 | int mem_cgroup_scan_tasks(struct mem_cgroup *memcg, | |
1074 | int (*fn)(struct task_struct *, void *), void *arg) | |
1075 | { | |
1076 | struct mem_cgroup *iter; | |
1077 | int ret = 0; | |
1078 | ||
1079 | BUG_ON(memcg == root_mem_cgroup); | |
1080 | ||
1081 | for_each_mem_cgroup_tree(iter, memcg) { | |
1082 | struct css_task_iter it; | |
1083 | struct task_struct *task; | |
1084 | ||
bc2fb7ed | 1085 | css_task_iter_start(&iter->css, 0, &it); |
7c5f64f8 VD |
1086 | while (!ret && (task = css_task_iter_next(&it))) |
1087 | ret = fn(task, arg); | |
1088 | css_task_iter_end(&it); | |
1089 | if (ret) { | |
1090 | mem_cgroup_iter_break(memcg, iter); | |
1091 | break; | |
1092 | } | |
1093 | } | |
1094 | return ret; | |
1095 | } | |
1096 | ||
925b7673 | 1097 | /** |
dfe0e773 | 1098 | * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page |
925b7673 | 1099 | * @page: the page |
f144c390 | 1100 | * @pgdat: pgdat of the page |
dfe0e773 JW |
1101 | * |
1102 | * This function is only safe when following the LRU page isolation | |
1103 | * and putback protocol: the LRU lock must be held, and the page must | |
1104 | * either be PageLRU() or the caller must have isolated/allocated it. | |
925b7673 | 1105 | */ |
599d0c95 | 1106 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct pglist_data *pgdat) |
08e552c6 | 1107 | { |
ef8f2327 | 1108 | struct mem_cgroup_per_node *mz; |
925b7673 | 1109 | struct mem_cgroup *memcg; |
bea8c150 | 1110 | struct lruvec *lruvec; |
6d12e2d8 | 1111 | |
bea8c150 | 1112 | if (mem_cgroup_disabled()) { |
599d0c95 | 1113 | lruvec = &pgdat->lruvec; |
bea8c150 HD |
1114 | goto out; |
1115 | } | |
925b7673 | 1116 | |
1306a85a | 1117 | memcg = page->mem_cgroup; |
7512102c | 1118 | /* |
dfe0e773 | 1119 | * Swapcache readahead pages are added to the LRU - and |
29833315 | 1120 | * possibly migrated - before they are charged. |
7512102c | 1121 | */ |
29833315 JW |
1122 | if (!memcg) |
1123 | memcg = root_mem_cgroup; | |
7512102c | 1124 | |
ef8f2327 | 1125 | mz = mem_cgroup_page_nodeinfo(memcg, page); |
bea8c150 HD |
1126 | lruvec = &mz->lruvec; |
1127 | out: | |
1128 | /* | |
1129 | * Since a node can be onlined after the mem_cgroup was created, | |
1130 | * we have to be prepared to initialize lruvec->zone here; | |
1131 | * and if offlined then reonlined, we need to reinitialize it. | |
1132 | */ | |
599d0c95 MG |
1133 | if (unlikely(lruvec->pgdat != pgdat)) |
1134 | lruvec->pgdat = pgdat; | |
bea8c150 | 1135 | return lruvec; |
08e552c6 | 1136 | } |
b69408e8 | 1137 | |
925b7673 | 1138 | /** |
fa9add64 HD |
1139 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1140 | * @lruvec: mem_cgroup per zone lru vector | |
1141 | * @lru: index of lru list the page is sitting on | |
b4536f0c | 1142 | * @zid: zone id of the accounted pages |
fa9add64 | 1143 | * @nr_pages: positive when adding or negative when removing |
925b7673 | 1144 | * |
ca707239 HD |
1145 | * This function must be called under lru_lock, just before a page is added |
1146 | * to or just after a page is removed from an lru list (that ordering being | |
1147 | * so as to allow it to check that lru_size 0 is consistent with list_empty). | |
3f58a829 | 1148 | */ |
fa9add64 | 1149 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
b4536f0c | 1150 | int zid, int nr_pages) |
3f58a829 | 1151 | { |
ef8f2327 | 1152 | struct mem_cgroup_per_node *mz; |
fa9add64 | 1153 | unsigned long *lru_size; |
ca707239 | 1154 | long size; |
3f58a829 MK |
1155 | |
1156 | if (mem_cgroup_disabled()) | |
1157 | return; | |
1158 | ||
ef8f2327 | 1159 | mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec); |
b4536f0c | 1160 | lru_size = &mz->lru_zone_size[zid][lru]; |
ca707239 HD |
1161 | |
1162 | if (nr_pages < 0) | |
1163 | *lru_size += nr_pages; | |
1164 | ||
1165 | size = *lru_size; | |
b4536f0c MH |
1166 | if (WARN_ONCE(size < 0, |
1167 | "%s(%p, %d, %d): lru_size %ld\n", | |
1168 | __func__, lruvec, lru, nr_pages, size)) { | |
ca707239 HD |
1169 | VM_BUG_ON(1); |
1170 | *lru_size = 0; | |
1171 | } | |
1172 | ||
1173 | if (nr_pages > 0) | |
1174 | *lru_size += nr_pages; | |
08e552c6 | 1175 | } |
544122e5 | 1176 | |
2314b42d | 1177 | bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg) |
c3ac9a8a | 1178 | { |
2314b42d | 1179 | struct mem_cgroup *task_memcg; |
158e0a2d | 1180 | struct task_struct *p; |
ffbdccf5 | 1181 | bool ret; |
4c4a2214 | 1182 | |
158e0a2d | 1183 | p = find_lock_task_mm(task); |
de077d22 | 1184 | if (p) { |
2314b42d | 1185 | task_memcg = get_mem_cgroup_from_mm(p->mm); |
de077d22 DR |
1186 | task_unlock(p); |
1187 | } else { | |
1188 | /* | |
1189 | * All threads may have already detached their mm's, but the oom | |
1190 | * killer still needs to detect if they have already been oom | |
1191 | * killed to prevent needlessly killing additional tasks. | |
1192 | */ | |
ffbdccf5 | 1193 | rcu_read_lock(); |
2314b42d JW |
1194 | task_memcg = mem_cgroup_from_task(task); |
1195 | css_get(&task_memcg->css); | |
ffbdccf5 | 1196 | rcu_read_unlock(); |
de077d22 | 1197 | } |
2314b42d JW |
1198 | ret = mem_cgroup_is_descendant(task_memcg, memcg); |
1199 | css_put(&task_memcg->css); | |
4c4a2214 DR |
1200 | return ret; |
1201 | } | |
1202 | ||
19942822 | 1203 | /** |
9d11ea9f | 1204 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1205 | * @memcg: the memory cgroup |
19942822 | 1206 | * |
9d11ea9f | 1207 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1208 | * pages. |
19942822 | 1209 | */ |
c0ff4b85 | 1210 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1211 | { |
3e32cb2e JW |
1212 | unsigned long margin = 0; |
1213 | unsigned long count; | |
1214 | unsigned long limit; | |
9d11ea9f | 1215 | |
3e32cb2e | 1216 | count = page_counter_read(&memcg->memory); |
bbec2e15 | 1217 | limit = READ_ONCE(memcg->memory.max); |
3e32cb2e JW |
1218 | if (count < limit) |
1219 | margin = limit - count; | |
1220 | ||
7941d214 | 1221 | if (do_memsw_account()) { |
3e32cb2e | 1222 | count = page_counter_read(&memcg->memsw); |
bbec2e15 | 1223 | limit = READ_ONCE(memcg->memsw.max); |
3e32cb2e JW |
1224 | if (count <= limit) |
1225 | margin = min(margin, limit - count); | |
cbedbac3 LR |
1226 | else |
1227 | margin = 0; | |
3e32cb2e JW |
1228 | } |
1229 | ||
1230 | return margin; | |
19942822 JW |
1231 | } |
1232 | ||
32047e2a | 1233 | /* |
bdcbb659 | 1234 | * A routine for checking "mem" is under move_account() or not. |
32047e2a | 1235 | * |
bdcbb659 QH |
1236 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
1237 | * moving cgroups. This is for waiting at high-memory pressure | |
1238 | * caused by "move". | |
32047e2a | 1239 | */ |
c0ff4b85 | 1240 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1241 | { |
2bd9bb20 KH |
1242 | struct mem_cgroup *from; |
1243 | struct mem_cgroup *to; | |
4b534334 | 1244 | bool ret = false; |
2bd9bb20 KH |
1245 | /* |
1246 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1247 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1248 | */ | |
1249 | spin_lock(&mc.lock); | |
1250 | from = mc.from; | |
1251 | to = mc.to; | |
1252 | if (!from) | |
1253 | goto unlock; | |
3e92041d | 1254 | |
2314b42d JW |
1255 | ret = mem_cgroup_is_descendant(from, memcg) || |
1256 | mem_cgroup_is_descendant(to, memcg); | |
2bd9bb20 KH |
1257 | unlock: |
1258 | spin_unlock(&mc.lock); | |
4b534334 KH |
1259 | return ret; |
1260 | } | |
1261 | ||
c0ff4b85 | 1262 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1263 | { |
1264 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1265 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1266 | DEFINE_WAIT(wait); |
1267 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1268 | /* moving charge context might have finished. */ | |
1269 | if (mc.moving_task) | |
1270 | schedule(); | |
1271 | finish_wait(&mc.waitq, &wait); | |
1272 | return true; | |
1273 | } | |
1274 | } | |
1275 | return false; | |
1276 | } | |
1277 | ||
8ad6e404 | 1278 | static const unsigned int memcg1_stats[] = { |
71cd3113 JW |
1279 | MEMCG_CACHE, |
1280 | MEMCG_RSS, | |
1281 | MEMCG_RSS_HUGE, | |
1282 | NR_SHMEM, | |
1283 | NR_FILE_MAPPED, | |
1284 | NR_FILE_DIRTY, | |
1285 | NR_WRITEBACK, | |
1286 | MEMCG_SWAP, | |
1287 | }; | |
1288 | ||
1289 | static const char *const memcg1_stat_names[] = { | |
1290 | "cache", | |
1291 | "rss", | |
1292 | "rss_huge", | |
1293 | "shmem", | |
1294 | "mapped_file", | |
1295 | "dirty", | |
1296 | "writeback", | |
1297 | "swap", | |
1298 | }; | |
1299 | ||
58cf188e | 1300 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1301 | /** |
f0c867d9 | 1302 | * mem_cgroup_print_oom_context: Print OOM information relevant to |
1303 | * memory controller. | |
e222432b BS |
1304 | * @memcg: The memory cgroup that went over limit |
1305 | * @p: Task that is going to be killed | |
1306 | * | |
1307 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1308 | * enabled | |
1309 | */ | |
f0c867d9 | 1310 | void mem_cgroup_print_oom_context(struct mem_cgroup *memcg, struct task_struct *p) |
e222432b | 1311 | { |
e222432b BS |
1312 | rcu_read_lock(); |
1313 | ||
f0c867d9 | 1314 | if (memcg) { |
1315 | pr_cont(",oom_memcg="); | |
1316 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1317 | } else | |
1318 | pr_cont(",global_oom"); | |
2415b9f5 | 1319 | if (p) { |
f0c867d9 | 1320 | pr_cont(",task_memcg="); |
2415b9f5 | 1321 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); |
2415b9f5 | 1322 | } |
e222432b | 1323 | rcu_read_unlock(); |
f0c867d9 | 1324 | } |
1325 | ||
1326 | /** | |
1327 | * mem_cgroup_print_oom_meminfo: Print OOM memory information relevant to | |
1328 | * memory controller. | |
1329 | * @memcg: The memory cgroup that went over limit | |
1330 | */ | |
1331 | void mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg) | |
1332 | { | |
1333 | struct mem_cgroup *iter; | |
1334 | unsigned int i; | |
e222432b | 1335 | |
3e32cb2e JW |
1336 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
1337 | K((u64)page_counter_read(&memcg->memory)), | |
bbec2e15 | 1338 | K((u64)memcg->memory.max), memcg->memory.failcnt); |
3e32cb2e JW |
1339 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", |
1340 | K((u64)page_counter_read(&memcg->memsw)), | |
bbec2e15 | 1341 | K((u64)memcg->memsw.max), memcg->memsw.failcnt); |
3e32cb2e JW |
1342 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", |
1343 | K((u64)page_counter_read(&memcg->kmem)), | |
bbec2e15 | 1344 | K((u64)memcg->kmem.max), memcg->kmem.failcnt); |
58cf188e SZ |
1345 | |
1346 | for_each_mem_cgroup_tree(iter, memcg) { | |
e61734c5 TH |
1347 | pr_info("Memory cgroup stats for "); |
1348 | pr_cont_cgroup_path(iter->css.cgroup); | |
58cf188e SZ |
1349 | pr_cont(":"); |
1350 | ||
71cd3113 JW |
1351 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
1352 | if (memcg1_stats[i] == MEMCG_SWAP && !do_swap_account) | |
58cf188e | 1353 | continue; |
71cd3113 | 1354 | pr_cont(" %s:%luKB", memcg1_stat_names[i], |
ccda7f43 | 1355 | K(memcg_page_state(iter, memcg1_stats[i]))); |
58cf188e SZ |
1356 | } |
1357 | ||
1358 | for (i = 0; i < NR_LRU_LISTS; i++) | |
1359 | pr_cont(" %s:%luKB", mem_cgroup_lru_names[i], | |
1360 | K(mem_cgroup_nr_lru_pages(iter, BIT(i)))); | |
1361 | ||
1362 | pr_cont("\n"); | |
1363 | } | |
e222432b BS |
1364 | } |
1365 | ||
a63d83f4 DR |
1366 | /* |
1367 | * Return the memory (and swap, if configured) limit for a memcg. | |
1368 | */ | |
bbec2e15 | 1369 | unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg) |
a63d83f4 | 1370 | { |
bbec2e15 | 1371 | unsigned long max; |
f3e8eb70 | 1372 | |
bbec2e15 | 1373 | max = memcg->memory.max; |
9a5a8f19 | 1374 | if (mem_cgroup_swappiness(memcg)) { |
bbec2e15 RG |
1375 | unsigned long memsw_max; |
1376 | unsigned long swap_max; | |
9a5a8f19 | 1377 | |
bbec2e15 RG |
1378 | memsw_max = memcg->memsw.max; |
1379 | swap_max = memcg->swap.max; | |
1380 | swap_max = min(swap_max, (unsigned long)total_swap_pages); | |
1381 | max = min(max + swap_max, memsw_max); | |
9a5a8f19 | 1382 | } |
bbec2e15 | 1383 | return max; |
a63d83f4 DR |
1384 | } |
1385 | ||
b6e6edcf | 1386 | static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
19965460 | 1387 | int order) |
9cbb78bb | 1388 | { |
6e0fc46d DR |
1389 | struct oom_control oc = { |
1390 | .zonelist = NULL, | |
1391 | .nodemask = NULL, | |
2a966b77 | 1392 | .memcg = memcg, |
6e0fc46d DR |
1393 | .gfp_mask = gfp_mask, |
1394 | .order = order, | |
6e0fc46d | 1395 | }; |
7c5f64f8 | 1396 | bool ret; |
9cbb78bb | 1397 | |
7775face TH |
1398 | if (mutex_lock_killable(&oom_lock)) |
1399 | return true; | |
1400 | /* | |
1401 | * A few threads which were not waiting at mutex_lock_killable() can | |
1402 | * fail to bail out. Therefore, check again after holding oom_lock. | |
1403 | */ | |
1404 | ret = should_force_charge() || out_of_memory(&oc); | |
dc56401f | 1405 | mutex_unlock(&oom_lock); |
7c5f64f8 | 1406 | return ret; |
9cbb78bb DR |
1407 | } |
1408 | ||
ae6e71d3 MC |
1409 | #if MAX_NUMNODES > 1 |
1410 | ||
4d0c066d KH |
1411 | /** |
1412 | * test_mem_cgroup_node_reclaimable | |
dad7557e | 1413 | * @memcg: the target memcg |
4d0c066d KH |
1414 | * @nid: the node ID to be checked. |
1415 | * @noswap : specify true here if the user wants flle only information. | |
1416 | * | |
1417 | * This function returns whether the specified memcg contains any | |
1418 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1419 | * pages in the node. | |
1420 | */ | |
c0ff4b85 | 1421 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1422 | int nid, bool noswap) |
1423 | { | |
c0ff4b85 | 1424 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
4d0c066d KH |
1425 | return true; |
1426 | if (noswap || !total_swap_pages) | |
1427 | return false; | |
c0ff4b85 | 1428 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
4d0c066d KH |
1429 | return true; |
1430 | return false; | |
1431 | ||
1432 | } | |
889976db YH |
1433 | |
1434 | /* | |
1435 | * Always updating the nodemask is not very good - even if we have an empty | |
1436 | * list or the wrong list here, we can start from some node and traverse all | |
1437 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1438 | * | |
1439 | */ | |
c0ff4b85 | 1440 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1441 | { |
1442 | int nid; | |
453a9bf3 KH |
1443 | /* |
1444 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1445 | * pagein/pageout changes since the last update. | |
1446 | */ | |
c0ff4b85 | 1447 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1448 | return; |
c0ff4b85 | 1449 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1450 | return; |
1451 | ||
889976db | 1452 | /* make a nodemask where this memcg uses memory from */ |
31aaea4a | 1453 | memcg->scan_nodes = node_states[N_MEMORY]; |
889976db | 1454 | |
31aaea4a | 1455 | for_each_node_mask(nid, node_states[N_MEMORY]) { |
889976db | 1456 | |
c0ff4b85 R |
1457 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1458 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1459 | } |
453a9bf3 | 1460 | |
c0ff4b85 R |
1461 | atomic_set(&memcg->numainfo_events, 0); |
1462 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1463 | } |
1464 | ||
1465 | /* | |
1466 | * Selecting a node where we start reclaim from. Because what we need is just | |
1467 | * reducing usage counter, start from anywhere is O,K. Considering | |
1468 | * memory reclaim from current node, there are pros. and cons. | |
1469 | * | |
1470 | * Freeing memory from current node means freeing memory from a node which | |
1471 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1472 | * hit limits, it will see a contention on a node. But freeing from remote | |
1473 | * node means more costs for memory reclaim because of memory latency. | |
1474 | * | |
1475 | * Now, we use round-robin. Better algorithm is welcomed. | |
1476 | */ | |
c0ff4b85 | 1477 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1478 | { |
1479 | int node; | |
1480 | ||
c0ff4b85 R |
1481 | mem_cgroup_may_update_nodemask(memcg); |
1482 | node = memcg->last_scanned_node; | |
889976db | 1483 | |
0edaf86c | 1484 | node = next_node_in(node, memcg->scan_nodes); |
889976db | 1485 | /* |
fda3d69b MH |
1486 | * mem_cgroup_may_update_nodemask might have seen no reclaimmable pages |
1487 | * last time it really checked all the LRUs due to rate limiting. | |
1488 | * Fallback to the current node in that case for simplicity. | |
889976db YH |
1489 | */ |
1490 | if (unlikely(node == MAX_NUMNODES)) | |
1491 | node = numa_node_id(); | |
1492 | ||
c0ff4b85 | 1493 | memcg->last_scanned_node = node; |
889976db YH |
1494 | return node; |
1495 | } | |
889976db | 1496 | #else |
c0ff4b85 | 1497 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1498 | { |
1499 | return 0; | |
1500 | } | |
1501 | #endif | |
1502 | ||
0608f43d | 1503 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
ef8f2327 | 1504 | pg_data_t *pgdat, |
0608f43d AM |
1505 | gfp_t gfp_mask, |
1506 | unsigned long *total_scanned) | |
1507 | { | |
1508 | struct mem_cgroup *victim = NULL; | |
1509 | int total = 0; | |
1510 | int loop = 0; | |
1511 | unsigned long excess; | |
1512 | unsigned long nr_scanned; | |
1513 | struct mem_cgroup_reclaim_cookie reclaim = { | |
ef8f2327 | 1514 | .pgdat = pgdat, |
0608f43d AM |
1515 | .priority = 0, |
1516 | }; | |
1517 | ||
3e32cb2e | 1518 | excess = soft_limit_excess(root_memcg); |
0608f43d AM |
1519 | |
1520 | while (1) { | |
1521 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
1522 | if (!victim) { | |
1523 | loop++; | |
1524 | if (loop >= 2) { | |
1525 | /* | |
1526 | * If we have not been able to reclaim | |
1527 | * anything, it might because there are | |
1528 | * no reclaimable pages under this hierarchy | |
1529 | */ | |
1530 | if (!total) | |
1531 | break; | |
1532 | /* | |
1533 | * We want to do more targeted reclaim. | |
1534 | * excess >> 2 is not to excessive so as to | |
1535 | * reclaim too much, nor too less that we keep | |
1536 | * coming back to reclaim from this cgroup | |
1537 | */ | |
1538 | if (total >= (excess >> 2) || | |
1539 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
1540 | break; | |
1541 | } | |
1542 | continue; | |
1543 | } | |
a9dd0a83 | 1544 | total += mem_cgroup_shrink_node(victim, gfp_mask, false, |
ef8f2327 | 1545 | pgdat, &nr_scanned); |
0608f43d | 1546 | *total_scanned += nr_scanned; |
3e32cb2e | 1547 | if (!soft_limit_excess(root_memcg)) |
0608f43d | 1548 | break; |
6d61ef40 | 1549 | } |
0608f43d AM |
1550 | mem_cgroup_iter_break(root_memcg, victim); |
1551 | return total; | |
6d61ef40 BS |
1552 | } |
1553 | ||
0056f4e6 JW |
1554 | #ifdef CONFIG_LOCKDEP |
1555 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
1556 | .name = "memcg_oom_lock", | |
1557 | }; | |
1558 | #endif | |
1559 | ||
fb2a6fc5 JW |
1560 | static DEFINE_SPINLOCK(memcg_oom_lock); |
1561 | ||
867578cb KH |
1562 | /* |
1563 | * Check OOM-Killer is already running under our hierarchy. | |
1564 | * If someone is running, return false. | |
1565 | */ | |
fb2a6fc5 | 1566 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 1567 | { |
79dfdacc | 1568 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1569 | |
fb2a6fc5 JW |
1570 | spin_lock(&memcg_oom_lock); |
1571 | ||
9f3a0d09 | 1572 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1573 | if (iter->oom_lock) { |
79dfdacc MH |
1574 | /* |
1575 | * this subtree of our hierarchy is already locked | |
1576 | * so we cannot give a lock. | |
1577 | */ | |
79dfdacc | 1578 | failed = iter; |
9f3a0d09 JW |
1579 | mem_cgroup_iter_break(memcg, iter); |
1580 | break; | |
23751be0 JW |
1581 | } else |
1582 | iter->oom_lock = true; | |
7d74b06f | 1583 | } |
867578cb | 1584 | |
fb2a6fc5 JW |
1585 | if (failed) { |
1586 | /* | |
1587 | * OK, we failed to lock the whole subtree so we have | |
1588 | * to clean up what we set up to the failing subtree | |
1589 | */ | |
1590 | for_each_mem_cgroup_tree(iter, memcg) { | |
1591 | if (iter == failed) { | |
1592 | mem_cgroup_iter_break(memcg, iter); | |
1593 | break; | |
1594 | } | |
1595 | iter->oom_lock = false; | |
79dfdacc | 1596 | } |
0056f4e6 JW |
1597 | } else |
1598 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
1599 | |
1600 | spin_unlock(&memcg_oom_lock); | |
1601 | ||
1602 | return !failed; | |
a636b327 | 1603 | } |
0b7f569e | 1604 | |
fb2a6fc5 | 1605 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1606 | { |
7d74b06f KH |
1607 | struct mem_cgroup *iter; |
1608 | ||
fb2a6fc5 | 1609 | spin_lock(&memcg_oom_lock); |
0056f4e6 | 1610 | mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_); |
c0ff4b85 | 1611 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1612 | iter->oom_lock = false; |
fb2a6fc5 | 1613 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
1614 | } |
1615 | ||
c0ff4b85 | 1616 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1617 | { |
1618 | struct mem_cgroup *iter; | |
1619 | ||
c2b42d3c | 1620 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1621 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1622 | iter->under_oom++; |
1623 | spin_unlock(&memcg_oom_lock); | |
79dfdacc MH |
1624 | } |
1625 | ||
c0ff4b85 | 1626 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1627 | { |
1628 | struct mem_cgroup *iter; | |
1629 | ||
867578cb KH |
1630 | /* |
1631 | * When a new child is created while the hierarchy is under oom, | |
c2b42d3c | 1632 | * mem_cgroup_oom_lock() may not be called. Watch for underflow. |
867578cb | 1633 | */ |
c2b42d3c | 1634 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1635 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1636 | if (iter->under_oom > 0) |
1637 | iter->under_oom--; | |
1638 | spin_unlock(&memcg_oom_lock); | |
0b7f569e KH |
1639 | } |
1640 | ||
867578cb KH |
1641 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1642 | ||
dc98df5a | 1643 | struct oom_wait_info { |
d79154bb | 1644 | struct mem_cgroup *memcg; |
ac6424b9 | 1645 | wait_queue_entry_t wait; |
dc98df5a KH |
1646 | }; |
1647 | ||
ac6424b9 | 1648 | static int memcg_oom_wake_function(wait_queue_entry_t *wait, |
dc98df5a KH |
1649 | unsigned mode, int sync, void *arg) |
1650 | { | |
d79154bb HD |
1651 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
1652 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
1653 | struct oom_wait_info *oom_wait_info; |
1654 | ||
1655 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 1656 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 1657 | |
2314b42d JW |
1658 | if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) && |
1659 | !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg)) | |
dc98df5a | 1660 | return 0; |
dc98df5a KH |
1661 | return autoremove_wake_function(wait, mode, sync, arg); |
1662 | } | |
1663 | ||
c0ff4b85 | 1664 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1665 | { |
c2b42d3c TH |
1666 | /* |
1667 | * For the following lockless ->under_oom test, the only required | |
1668 | * guarantee is that it must see the state asserted by an OOM when | |
1669 | * this function is called as a result of userland actions | |
1670 | * triggered by the notification of the OOM. This is trivially | |
1671 | * achieved by invoking mem_cgroup_mark_under_oom() before | |
1672 | * triggering notification. | |
1673 | */ | |
1674 | if (memcg && memcg->under_oom) | |
f4b90b70 | 1675 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); |
3c11ecf4 KH |
1676 | } |
1677 | ||
29ef680a MH |
1678 | enum oom_status { |
1679 | OOM_SUCCESS, | |
1680 | OOM_FAILED, | |
1681 | OOM_ASYNC, | |
1682 | OOM_SKIPPED | |
1683 | }; | |
1684 | ||
1685 | static enum oom_status mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) | |
0b7f569e | 1686 | { |
7056d3a3 MH |
1687 | enum oom_status ret; |
1688 | bool locked; | |
1689 | ||
29ef680a MH |
1690 | if (order > PAGE_ALLOC_COSTLY_ORDER) |
1691 | return OOM_SKIPPED; | |
1692 | ||
7a1adfdd RG |
1693 | memcg_memory_event(memcg, MEMCG_OOM); |
1694 | ||
867578cb | 1695 | /* |
49426420 JW |
1696 | * We are in the middle of the charge context here, so we |
1697 | * don't want to block when potentially sitting on a callstack | |
1698 | * that holds all kinds of filesystem and mm locks. | |
1699 | * | |
29ef680a MH |
1700 | * cgroup1 allows disabling the OOM killer and waiting for outside |
1701 | * handling until the charge can succeed; remember the context and put | |
1702 | * the task to sleep at the end of the page fault when all locks are | |
1703 | * released. | |
49426420 | 1704 | * |
29ef680a MH |
1705 | * On the other hand, in-kernel OOM killer allows for an async victim |
1706 | * memory reclaim (oom_reaper) and that means that we are not solely | |
1707 | * relying on the oom victim to make a forward progress and we can | |
1708 | * invoke the oom killer here. | |
1709 | * | |
1710 | * Please note that mem_cgroup_out_of_memory might fail to find a | |
1711 | * victim and then we have to bail out from the charge path. | |
867578cb | 1712 | */ |
29ef680a MH |
1713 | if (memcg->oom_kill_disable) { |
1714 | if (!current->in_user_fault) | |
1715 | return OOM_SKIPPED; | |
1716 | css_get(&memcg->css); | |
1717 | current->memcg_in_oom = memcg; | |
1718 | current->memcg_oom_gfp_mask = mask; | |
1719 | current->memcg_oom_order = order; | |
1720 | ||
1721 | return OOM_ASYNC; | |
1722 | } | |
1723 | ||
7056d3a3 MH |
1724 | mem_cgroup_mark_under_oom(memcg); |
1725 | ||
1726 | locked = mem_cgroup_oom_trylock(memcg); | |
1727 | ||
1728 | if (locked) | |
1729 | mem_cgroup_oom_notify(memcg); | |
1730 | ||
1731 | mem_cgroup_unmark_under_oom(memcg); | |
29ef680a | 1732 | if (mem_cgroup_out_of_memory(memcg, mask, order)) |
7056d3a3 MH |
1733 | ret = OOM_SUCCESS; |
1734 | else | |
1735 | ret = OOM_FAILED; | |
1736 | ||
1737 | if (locked) | |
1738 | mem_cgroup_oom_unlock(memcg); | |
29ef680a | 1739 | |
7056d3a3 | 1740 | return ret; |
3812c8c8 JW |
1741 | } |
1742 | ||
1743 | /** | |
1744 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 1745 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 1746 | * |
49426420 JW |
1747 | * This has to be called at the end of a page fault if the memcg OOM |
1748 | * handler was enabled. | |
3812c8c8 | 1749 | * |
49426420 | 1750 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
1751 | * sleep on a waitqueue until the userspace task resolves the |
1752 | * situation. Sleeping directly in the charge context with all kinds | |
1753 | * of locks held is not a good idea, instead we remember an OOM state | |
1754 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 1755 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
1756 | * |
1757 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 1758 | * completed, %false otherwise. |
3812c8c8 | 1759 | */ |
49426420 | 1760 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 1761 | { |
626ebc41 | 1762 | struct mem_cgroup *memcg = current->memcg_in_oom; |
3812c8c8 | 1763 | struct oom_wait_info owait; |
49426420 | 1764 | bool locked; |
3812c8c8 JW |
1765 | |
1766 | /* OOM is global, do not handle */ | |
3812c8c8 | 1767 | if (!memcg) |
49426420 | 1768 | return false; |
3812c8c8 | 1769 | |
7c5f64f8 | 1770 | if (!handle) |
49426420 | 1771 | goto cleanup; |
3812c8c8 JW |
1772 | |
1773 | owait.memcg = memcg; | |
1774 | owait.wait.flags = 0; | |
1775 | owait.wait.func = memcg_oom_wake_function; | |
1776 | owait.wait.private = current; | |
2055da97 | 1777 | INIT_LIST_HEAD(&owait.wait.entry); |
867578cb | 1778 | |
3812c8c8 | 1779 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
1780 | mem_cgroup_mark_under_oom(memcg); |
1781 | ||
1782 | locked = mem_cgroup_oom_trylock(memcg); | |
1783 | ||
1784 | if (locked) | |
1785 | mem_cgroup_oom_notify(memcg); | |
1786 | ||
1787 | if (locked && !memcg->oom_kill_disable) { | |
1788 | mem_cgroup_unmark_under_oom(memcg); | |
1789 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
626ebc41 TH |
1790 | mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask, |
1791 | current->memcg_oom_order); | |
49426420 | 1792 | } else { |
3812c8c8 | 1793 | schedule(); |
49426420 JW |
1794 | mem_cgroup_unmark_under_oom(memcg); |
1795 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
1796 | } | |
1797 | ||
1798 | if (locked) { | |
fb2a6fc5 JW |
1799 | mem_cgroup_oom_unlock(memcg); |
1800 | /* | |
1801 | * There is no guarantee that an OOM-lock contender | |
1802 | * sees the wakeups triggered by the OOM kill | |
1803 | * uncharges. Wake any sleepers explicitely. | |
1804 | */ | |
1805 | memcg_oom_recover(memcg); | |
1806 | } | |
49426420 | 1807 | cleanup: |
626ebc41 | 1808 | current->memcg_in_oom = NULL; |
3812c8c8 | 1809 | css_put(&memcg->css); |
867578cb | 1810 | return true; |
0b7f569e KH |
1811 | } |
1812 | ||
3d8b38eb RG |
1813 | /** |
1814 | * mem_cgroup_get_oom_group - get a memory cgroup to clean up after OOM | |
1815 | * @victim: task to be killed by the OOM killer | |
1816 | * @oom_domain: memcg in case of memcg OOM, NULL in case of system-wide OOM | |
1817 | * | |
1818 | * Returns a pointer to a memory cgroup, which has to be cleaned up | |
1819 | * by killing all belonging OOM-killable tasks. | |
1820 | * | |
1821 | * Caller has to call mem_cgroup_put() on the returned non-NULL memcg. | |
1822 | */ | |
1823 | struct mem_cgroup *mem_cgroup_get_oom_group(struct task_struct *victim, | |
1824 | struct mem_cgroup *oom_domain) | |
1825 | { | |
1826 | struct mem_cgroup *oom_group = NULL; | |
1827 | struct mem_cgroup *memcg; | |
1828 | ||
1829 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
1830 | return NULL; | |
1831 | ||
1832 | if (!oom_domain) | |
1833 | oom_domain = root_mem_cgroup; | |
1834 | ||
1835 | rcu_read_lock(); | |
1836 | ||
1837 | memcg = mem_cgroup_from_task(victim); | |
1838 | if (memcg == root_mem_cgroup) | |
1839 | goto out; | |
1840 | ||
1841 | /* | |
1842 | * Traverse the memory cgroup hierarchy from the victim task's | |
1843 | * cgroup up to the OOMing cgroup (or root) to find the | |
1844 | * highest-level memory cgroup with oom.group set. | |
1845 | */ | |
1846 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
1847 | if (memcg->oom_group) | |
1848 | oom_group = memcg; | |
1849 | ||
1850 | if (memcg == oom_domain) | |
1851 | break; | |
1852 | } | |
1853 | ||
1854 | if (oom_group) | |
1855 | css_get(&oom_group->css); | |
1856 | out: | |
1857 | rcu_read_unlock(); | |
1858 | ||
1859 | return oom_group; | |
1860 | } | |
1861 | ||
1862 | void mem_cgroup_print_oom_group(struct mem_cgroup *memcg) | |
1863 | { | |
1864 | pr_info("Tasks in "); | |
1865 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1866 | pr_cont(" are going to be killed due to memory.oom.group set\n"); | |
1867 | } | |
1868 | ||
d7365e78 | 1869 | /** |
81f8c3a4 JW |
1870 | * lock_page_memcg - lock a page->mem_cgroup binding |
1871 | * @page: the page | |
32047e2a | 1872 | * |
81f8c3a4 | 1873 | * This function protects unlocked LRU pages from being moved to |
739f79fc JW |
1874 | * another cgroup. |
1875 | * | |
1876 | * It ensures lifetime of the returned memcg. Caller is responsible | |
1877 | * for the lifetime of the page; __unlock_page_memcg() is available | |
1878 | * when @page might get freed inside the locked section. | |
d69b042f | 1879 | */ |
739f79fc | 1880 | struct mem_cgroup *lock_page_memcg(struct page *page) |
89c06bd5 KH |
1881 | { |
1882 | struct mem_cgroup *memcg; | |
6de22619 | 1883 | unsigned long flags; |
89c06bd5 | 1884 | |
6de22619 JW |
1885 | /* |
1886 | * The RCU lock is held throughout the transaction. The fast | |
1887 | * path can get away without acquiring the memcg->move_lock | |
1888 | * because page moving starts with an RCU grace period. | |
739f79fc JW |
1889 | * |
1890 | * The RCU lock also protects the memcg from being freed when | |
1891 | * the page state that is going to change is the only thing | |
1892 | * preventing the page itself from being freed. E.g. writeback | |
1893 | * doesn't hold a page reference and relies on PG_writeback to | |
1894 | * keep off truncation, migration and so forth. | |
1895 | */ | |
d7365e78 JW |
1896 | rcu_read_lock(); |
1897 | ||
1898 | if (mem_cgroup_disabled()) | |
739f79fc | 1899 | return NULL; |
89c06bd5 | 1900 | again: |
1306a85a | 1901 | memcg = page->mem_cgroup; |
29833315 | 1902 | if (unlikely(!memcg)) |
739f79fc | 1903 | return NULL; |
d7365e78 | 1904 | |
bdcbb659 | 1905 | if (atomic_read(&memcg->moving_account) <= 0) |
739f79fc | 1906 | return memcg; |
89c06bd5 | 1907 | |
6de22619 | 1908 | spin_lock_irqsave(&memcg->move_lock, flags); |
1306a85a | 1909 | if (memcg != page->mem_cgroup) { |
6de22619 | 1910 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
89c06bd5 KH |
1911 | goto again; |
1912 | } | |
6de22619 JW |
1913 | |
1914 | /* | |
1915 | * When charge migration first begins, we can have locked and | |
1916 | * unlocked page stat updates happening concurrently. Track | |
81f8c3a4 | 1917 | * the task who has the lock for unlock_page_memcg(). |
6de22619 JW |
1918 | */ |
1919 | memcg->move_lock_task = current; | |
1920 | memcg->move_lock_flags = flags; | |
d7365e78 | 1921 | |
739f79fc | 1922 | return memcg; |
89c06bd5 | 1923 | } |
81f8c3a4 | 1924 | EXPORT_SYMBOL(lock_page_memcg); |
89c06bd5 | 1925 | |
d7365e78 | 1926 | /** |
739f79fc JW |
1927 | * __unlock_page_memcg - unlock and unpin a memcg |
1928 | * @memcg: the memcg | |
1929 | * | |
1930 | * Unlock and unpin a memcg returned by lock_page_memcg(). | |
d7365e78 | 1931 | */ |
739f79fc | 1932 | void __unlock_page_memcg(struct mem_cgroup *memcg) |
89c06bd5 | 1933 | { |
6de22619 JW |
1934 | if (memcg && memcg->move_lock_task == current) { |
1935 | unsigned long flags = memcg->move_lock_flags; | |
1936 | ||
1937 | memcg->move_lock_task = NULL; | |
1938 | memcg->move_lock_flags = 0; | |
1939 | ||
1940 | spin_unlock_irqrestore(&memcg->move_lock, flags); | |
1941 | } | |
89c06bd5 | 1942 | |
d7365e78 | 1943 | rcu_read_unlock(); |
89c06bd5 | 1944 | } |
739f79fc JW |
1945 | |
1946 | /** | |
1947 | * unlock_page_memcg - unlock a page->mem_cgroup binding | |
1948 | * @page: the page | |
1949 | */ | |
1950 | void unlock_page_memcg(struct page *page) | |
1951 | { | |
1952 | __unlock_page_memcg(page->mem_cgroup); | |
1953 | } | |
81f8c3a4 | 1954 | EXPORT_SYMBOL(unlock_page_memcg); |
89c06bd5 | 1955 | |
cdec2e42 KH |
1956 | struct memcg_stock_pcp { |
1957 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 1958 | unsigned int nr_pages; |
cdec2e42 | 1959 | struct work_struct work; |
26fe6168 | 1960 | unsigned long flags; |
a0db00fc | 1961 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
1962 | }; |
1963 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 1964 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 1965 | |
a0956d54 SS |
1966 | /** |
1967 | * consume_stock: Try to consume stocked charge on this cpu. | |
1968 | * @memcg: memcg to consume from. | |
1969 | * @nr_pages: how many pages to charge. | |
1970 | * | |
1971 | * The charges will only happen if @memcg matches the current cpu's memcg | |
1972 | * stock, and at least @nr_pages are available in that stock. Failure to | |
1973 | * service an allocation will refill the stock. | |
1974 | * | |
1975 | * returns true if successful, false otherwise. | |
cdec2e42 | 1976 | */ |
a0956d54 | 1977 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
1978 | { |
1979 | struct memcg_stock_pcp *stock; | |
db2ba40c | 1980 | unsigned long flags; |
3e32cb2e | 1981 | bool ret = false; |
cdec2e42 | 1982 | |
a983b5eb | 1983 | if (nr_pages > MEMCG_CHARGE_BATCH) |
3e32cb2e | 1984 | return ret; |
a0956d54 | 1985 | |
db2ba40c JW |
1986 | local_irq_save(flags); |
1987 | ||
1988 | stock = this_cpu_ptr(&memcg_stock); | |
3e32cb2e | 1989 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) { |
a0956d54 | 1990 | stock->nr_pages -= nr_pages; |
3e32cb2e JW |
1991 | ret = true; |
1992 | } | |
db2ba40c JW |
1993 | |
1994 | local_irq_restore(flags); | |
1995 | ||
cdec2e42 KH |
1996 | return ret; |
1997 | } | |
1998 | ||
1999 | /* | |
3e32cb2e | 2000 | * Returns stocks cached in percpu and reset cached information. |
cdec2e42 KH |
2001 | */ |
2002 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2003 | { | |
2004 | struct mem_cgroup *old = stock->cached; | |
2005 | ||
11c9ea4e | 2006 | if (stock->nr_pages) { |
3e32cb2e | 2007 | page_counter_uncharge(&old->memory, stock->nr_pages); |
7941d214 | 2008 | if (do_memsw_account()) |
3e32cb2e | 2009 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
e8ea14cc | 2010 | css_put_many(&old->css, stock->nr_pages); |
11c9ea4e | 2011 | stock->nr_pages = 0; |
cdec2e42 KH |
2012 | } |
2013 | stock->cached = NULL; | |
cdec2e42 KH |
2014 | } |
2015 | ||
cdec2e42 KH |
2016 | static void drain_local_stock(struct work_struct *dummy) |
2017 | { | |
db2ba40c JW |
2018 | struct memcg_stock_pcp *stock; |
2019 | unsigned long flags; | |
2020 | ||
72f0184c MH |
2021 | /* |
2022 | * The only protection from memory hotplug vs. drain_stock races is | |
2023 | * that we always operate on local CPU stock here with IRQ disabled | |
2024 | */ | |
db2ba40c JW |
2025 | local_irq_save(flags); |
2026 | ||
2027 | stock = this_cpu_ptr(&memcg_stock); | |
cdec2e42 | 2028 | drain_stock(stock); |
26fe6168 | 2029 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
db2ba40c JW |
2030 | |
2031 | local_irq_restore(flags); | |
cdec2e42 KH |
2032 | } |
2033 | ||
2034 | /* | |
3e32cb2e | 2035 | * Cache charges(val) to local per_cpu area. |
320cc51d | 2036 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2037 | */ |
c0ff4b85 | 2038 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 | 2039 | { |
db2ba40c JW |
2040 | struct memcg_stock_pcp *stock; |
2041 | unsigned long flags; | |
2042 | ||
2043 | local_irq_save(flags); | |
cdec2e42 | 2044 | |
db2ba40c | 2045 | stock = this_cpu_ptr(&memcg_stock); |
c0ff4b85 | 2046 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2047 | drain_stock(stock); |
c0ff4b85 | 2048 | stock->cached = memcg; |
cdec2e42 | 2049 | } |
11c9ea4e | 2050 | stock->nr_pages += nr_pages; |
db2ba40c | 2051 | |
a983b5eb | 2052 | if (stock->nr_pages > MEMCG_CHARGE_BATCH) |
475d0487 RG |
2053 | drain_stock(stock); |
2054 | ||
db2ba40c | 2055 | local_irq_restore(flags); |
cdec2e42 KH |
2056 | } |
2057 | ||
2058 | /* | |
c0ff4b85 | 2059 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
6d3d6aa2 | 2060 | * of the hierarchy under it. |
cdec2e42 | 2061 | */ |
6d3d6aa2 | 2062 | static void drain_all_stock(struct mem_cgroup *root_memcg) |
cdec2e42 | 2063 | { |
26fe6168 | 2064 | int cpu, curcpu; |
d38144b7 | 2065 | |
6d3d6aa2 JW |
2066 | /* If someone's already draining, avoid adding running more workers. */ |
2067 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2068 | return; | |
72f0184c MH |
2069 | /* |
2070 | * Notify other cpus that system-wide "drain" is running | |
2071 | * We do not care about races with the cpu hotplug because cpu down | |
2072 | * as well as workers from this path always operate on the local | |
2073 | * per-cpu data. CPU up doesn't touch memcg_stock at all. | |
2074 | */ | |
5af12d0e | 2075 | curcpu = get_cpu(); |
cdec2e42 KH |
2076 | for_each_online_cpu(cpu) { |
2077 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2078 | struct mem_cgroup *memcg; |
26fe6168 | 2079 | |
c0ff4b85 | 2080 | memcg = stock->cached; |
72f0184c | 2081 | if (!memcg || !stock->nr_pages || !css_tryget(&memcg->css)) |
26fe6168 | 2082 | continue; |
72f0184c MH |
2083 | if (!mem_cgroup_is_descendant(memcg, root_memcg)) { |
2084 | css_put(&memcg->css); | |
3e92041d | 2085 | continue; |
72f0184c | 2086 | } |
d1a05b69 MH |
2087 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
2088 | if (cpu == curcpu) | |
2089 | drain_local_stock(&stock->work); | |
2090 | else | |
2091 | schedule_work_on(cpu, &stock->work); | |
2092 | } | |
72f0184c | 2093 | css_put(&memcg->css); |
cdec2e42 | 2094 | } |
5af12d0e | 2095 | put_cpu(); |
9f50fad6 | 2096 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2097 | } |
2098 | ||
308167fc | 2099 | static int memcg_hotplug_cpu_dead(unsigned int cpu) |
cdec2e42 | 2100 | { |
cdec2e42 | 2101 | struct memcg_stock_pcp *stock; |
a983b5eb | 2102 | struct mem_cgroup *memcg; |
cdec2e42 | 2103 | |
cdec2e42 KH |
2104 | stock = &per_cpu(memcg_stock, cpu); |
2105 | drain_stock(stock); | |
a983b5eb JW |
2106 | |
2107 | for_each_mem_cgroup(memcg) { | |
2108 | int i; | |
2109 | ||
2110 | for (i = 0; i < MEMCG_NR_STAT; i++) { | |
2111 | int nid; | |
2112 | long x; | |
2113 | ||
2114 | x = this_cpu_xchg(memcg->stat_cpu->count[i], 0); | |
2115 | if (x) | |
2116 | atomic_long_add(x, &memcg->stat[i]); | |
2117 | ||
2118 | if (i >= NR_VM_NODE_STAT_ITEMS) | |
2119 | continue; | |
2120 | ||
2121 | for_each_node(nid) { | |
2122 | struct mem_cgroup_per_node *pn; | |
2123 | ||
2124 | pn = mem_cgroup_nodeinfo(memcg, nid); | |
2125 | x = this_cpu_xchg(pn->lruvec_stat_cpu->count[i], 0); | |
2126 | if (x) | |
2127 | atomic_long_add(x, &pn->lruvec_stat[i]); | |
2128 | } | |
2129 | } | |
2130 | ||
e27be240 | 2131 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) { |
a983b5eb JW |
2132 | long x; |
2133 | ||
2134 | x = this_cpu_xchg(memcg->stat_cpu->events[i], 0); | |
2135 | if (x) | |
2136 | atomic_long_add(x, &memcg->events[i]); | |
2137 | } | |
2138 | } | |
2139 | ||
308167fc | 2140 | return 0; |
cdec2e42 KH |
2141 | } |
2142 | ||
f7e1cb6e JW |
2143 | static void reclaim_high(struct mem_cgroup *memcg, |
2144 | unsigned int nr_pages, | |
2145 | gfp_t gfp_mask) | |
2146 | { | |
2147 | do { | |
2148 | if (page_counter_read(&memcg->memory) <= memcg->high) | |
2149 | continue; | |
e27be240 | 2150 | memcg_memory_event(memcg, MEMCG_HIGH); |
f7e1cb6e JW |
2151 | try_to_free_mem_cgroup_pages(memcg, nr_pages, gfp_mask, true); |
2152 | } while ((memcg = parent_mem_cgroup(memcg))); | |
2153 | } | |
2154 | ||
2155 | static void high_work_func(struct work_struct *work) | |
2156 | { | |
2157 | struct mem_cgroup *memcg; | |
2158 | ||
2159 | memcg = container_of(work, struct mem_cgroup, high_work); | |
a983b5eb | 2160 | reclaim_high(memcg, MEMCG_CHARGE_BATCH, GFP_KERNEL); |
f7e1cb6e JW |
2161 | } |
2162 | ||
b23afb93 TH |
2163 | /* |
2164 | * Scheduled by try_charge() to be executed from the userland return path | |
2165 | * and reclaims memory over the high limit. | |
2166 | */ | |
2167 | void mem_cgroup_handle_over_high(void) | |
2168 | { | |
2169 | unsigned int nr_pages = current->memcg_nr_pages_over_high; | |
f7e1cb6e | 2170 | struct mem_cgroup *memcg; |
b23afb93 TH |
2171 | |
2172 | if (likely(!nr_pages)) | |
2173 | return; | |
2174 | ||
f7e1cb6e JW |
2175 | memcg = get_mem_cgroup_from_mm(current->mm); |
2176 | reclaim_high(memcg, nr_pages, GFP_KERNEL); | |
b23afb93 TH |
2177 | css_put(&memcg->css); |
2178 | current->memcg_nr_pages_over_high = 0; | |
2179 | } | |
2180 | ||
00501b53 JW |
2181 | static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
2182 | unsigned int nr_pages) | |
8a9f3ccd | 2183 | { |
a983b5eb | 2184 | unsigned int batch = max(MEMCG_CHARGE_BATCH, nr_pages); |
9b130619 | 2185 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
6539cc05 | 2186 | struct mem_cgroup *mem_over_limit; |
3e32cb2e | 2187 | struct page_counter *counter; |
6539cc05 | 2188 | unsigned long nr_reclaimed; |
b70a2a21 JW |
2189 | bool may_swap = true; |
2190 | bool drained = false; | |
29ef680a MH |
2191 | bool oomed = false; |
2192 | enum oom_status oom_status; | |
a636b327 | 2193 | |
ce00a967 | 2194 | if (mem_cgroup_is_root(memcg)) |
10d53c74 | 2195 | return 0; |
6539cc05 | 2196 | retry: |
b6b6cc72 | 2197 | if (consume_stock(memcg, nr_pages)) |
10d53c74 | 2198 | return 0; |
8a9f3ccd | 2199 | |
7941d214 | 2200 | if (!do_memsw_account() || |
6071ca52 JW |
2201 | page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
2202 | if (page_counter_try_charge(&memcg->memory, batch, &counter)) | |
6539cc05 | 2203 | goto done_restock; |
7941d214 | 2204 | if (do_memsw_account()) |
3e32cb2e JW |
2205 | page_counter_uncharge(&memcg->memsw, batch); |
2206 | mem_over_limit = mem_cgroup_from_counter(counter, memory); | |
3fbe7244 | 2207 | } else { |
3e32cb2e | 2208 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
b70a2a21 | 2209 | may_swap = false; |
3fbe7244 | 2210 | } |
7a81b88c | 2211 | |
6539cc05 JW |
2212 | if (batch > nr_pages) { |
2213 | batch = nr_pages; | |
2214 | goto retry; | |
2215 | } | |
6d61ef40 | 2216 | |
06b078fc JW |
2217 | /* |
2218 | * Unlike in global OOM situations, memcg is not in a physical | |
2219 | * memory shortage. Allow dying and OOM-killed tasks to | |
2220 | * bypass the last charges so that they can exit quickly and | |
2221 | * free their memory. | |
2222 | */ | |
7775face | 2223 | if (unlikely(should_force_charge())) |
10d53c74 | 2224 | goto force; |
06b078fc | 2225 | |
89a28483 JW |
2226 | /* |
2227 | * Prevent unbounded recursion when reclaim operations need to | |
2228 | * allocate memory. This might exceed the limits temporarily, | |
2229 | * but we prefer facilitating memory reclaim and getting back | |
2230 | * under the limit over triggering OOM kills in these cases. | |
2231 | */ | |
2232 | if (unlikely(current->flags & PF_MEMALLOC)) | |
2233 | goto force; | |
2234 | ||
06b078fc JW |
2235 | if (unlikely(task_in_memcg_oom(current))) |
2236 | goto nomem; | |
2237 | ||
d0164adc | 2238 | if (!gfpflags_allow_blocking(gfp_mask)) |
6539cc05 | 2239 | goto nomem; |
4b534334 | 2240 | |
e27be240 | 2241 | memcg_memory_event(mem_over_limit, MEMCG_MAX); |
241994ed | 2242 | |
b70a2a21 JW |
2243 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
2244 | gfp_mask, may_swap); | |
6539cc05 | 2245 | |
61e02c74 | 2246 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
6539cc05 | 2247 | goto retry; |
28c34c29 | 2248 | |
b70a2a21 | 2249 | if (!drained) { |
6d3d6aa2 | 2250 | drain_all_stock(mem_over_limit); |
b70a2a21 JW |
2251 | drained = true; |
2252 | goto retry; | |
2253 | } | |
2254 | ||
28c34c29 JW |
2255 | if (gfp_mask & __GFP_NORETRY) |
2256 | goto nomem; | |
6539cc05 JW |
2257 | /* |
2258 | * Even though the limit is exceeded at this point, reclaim | |
2259 | * may have been able to free some pages. Retry the charge | |
2260 | * before killing the task. | |
2261 | * | |
2262 | * Only for regular pages, though: huge pages are rather | |
2263 | * unlikely to succeed so close to the limit, and we fall back | |
2264 | * to regular pages anyway in case of failure. | |
2265 | */ | |
61e02c74 | 2266 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
6539cc05 JW |
2267 | goto retry; |
2268 | /* | |
2269 | * At task move, charge accounts can be doubly counted. So, it's | |
2270 | * better to wait until the end of task_move if something is going on. | |
2271 | */ | |
2272 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2273 | goto retry; | |
2274 | ||
9b130619 JW |
2275 | if (nr_retries--) |
2276 | goto retry; | |
2277 | ||
29ef680a MH |
2278 | if (gfp_mask & __GFP_RETRY_MAYFAIL && oomed) |
2279 | goto nomem; | |
2280 | ||
06b078fc | 2281 | if (gfp_mask & __GFP_NOFAIL) |
10d53c74 | 2282 | goto force; |
06b078fc | 2283 | |
6539cc05 | 2284 | if (fatal_signal_pending(current)) |
10d53c74 | 2285 | goto force; |
6539cc05 | 2286 | |
29ef680a MH |
2287 | /* |
2288 | * keep retrying as long as the memcg oom killer is able to make | |
2289 | * a forward progress or bypass the charge if the oom killer | |
2290 | * couldn't make any progress. | |
2291 | */ | |
2292 | oom_status = mem_cgroup_oom(mem_over_limit, gfp_mask, | |
3608de07 | 2293 | get_order(nr_pages * PAGE_SIZE)); |
29ef680a MH |
2294 | switch (oom_status) { |
2295 | case OOM_SUCCESS: | |
2296 | nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
2297 | oomed = true; | |
2298 | goto retry; | |
2299 | case OOM_FAILED: | |
2300 | goto force; | |
2301 | default: | |
2302 | goto nomem; | |
2303 | } | |
7a81b88c | 2304 | nomem: |
6d1fdc48 | 2305 | if (!(gfp_mask & __GFP_NOFAIL)) |
3168ecbe | 2306 | return -ENOMEM; |
10d53c74 TH |
2307 | force: |
2308 | /* | |
2309 | * The allocation either can't fail or will lead to more memory | |
2310 | * being freed very soon. Allow memory usage go over the limit | |
2311 | * temporarily by force charging it. | |
2312 | */ | |
2313 | page_counter_charge(&memcg->memory, nr_pages); | |
7941d214 | 2314 | if (do_memsw_account()) |
10d53c74 TH |
2315 | page_counter_charge(&memcg->memsw, nr_pages); |
2316 | css_get_many(&memcg->css, nr_pages); | |
2317 | ||
2318 | return 0; | |
6539cc05 JW |
2319 | |
2320 | done_restock: | |
e8ea14cc | 2321 | css_get_many(&memcg->css, batch); |
6539cc05 JW |
2322 | if (batch > nr_pages) |
2323 | refill_stock(memcg, batch - nr_pages); | |
b23afb93 | 2324 | |
241994ed | 2325 | /* |
b23afb93 TH |
2326 | * If the hierarchy is above the normal consumption range, schedule |
2327 | * reclaim on returning to userland. We can perform reclaim here | |
71baba4b | 2328 | * if __GFP_RECLAIM but let's always punt for simplicity and so that |
b23afb93 TH |
2329 | * GFP_KERNEL can consistently be used during reclaim. @memcg is |
2330 | * not recorded as it most likely matches current's and won't | |
2331 | * change in the meantime. As high limit is checked again before | |
2332 | * reclaim, the cost of mismatch is negligible. | |
241994ed JW |
2333 | */ |
2334 | do { | |
b23afb93 | 2335 | if (page_counter_read(&memcg->memory) > memcg->high) { |
f7e1cb6e JW |
2336 | /* Don't bother a random interrupted task */ |
2337 | if (in_interrupt()) { | |
2338 | schedule_work(&memcg->high_work); | |
2339 | break; | |
2340 | } | |
9516a18a | 2341 | current->memcg_nr_pages_over_high += batch; |
b23afb93 TH |
2342 | set_notify_resume(current); |
2343 | break; | |
2344 | } | |
241994ed | 2345 | } while ((memcg = parent_mem_cgroup(memcg))); |
10d53c74 TH |
2346 | |
2347 | return 0; | |
7a81b88c | 2348 | } |
8a9f3ccd | 2349 | |
00501b53 | 2350 | static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
a3032a2c | 2351 | { |
ce00a967 JW |
2352 | if (mem_cgroup_is_root(memcg)) |
2353 | return; | |
2354 | ||
3e32cb2e | 2355 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 2356 | if (do_memsw_account()) |
3e32cb2e | 2357 | page_counter_uncharge(&memcg->memsw, nr_pages); |
ce00a967 | 2358 | |
e8ea14cc | 2359 | css_put_many(&memcg->css, nr_pages); |
d01dd17f KH |
2360 | } |
2361 | ||
0a31bc97 JW |
2362 | static void lock_page_lru(struct page *page, int *isolated) |
2363 | { | |
2364 | struct zone *zone = page_zone(page); | |
2365 | ||
a52633d8 | 2366 | spin_lock_irq(zone_lru_lock(zone)); |
0a31bc97 JW |
2367 | if (PageLRU(page)) { |
2368 | struct lruvec *lruvec; | |
2369 | ||
599d0c95 | 2370 | lruvec = mem_cgroup_page_lruvec(page, zone->zone_pgdat); |
0a31bc97 JW |
2371 | ClearPageLRU(page); |
2372 | del_page_from_lru_list(page, lruvec, page_lru(page)); | |
2373 | *isolated = 1; | |
2374 | } else | |
2375 | *isolated = 0; | |
2376 | } | |
2377 | ||
2378 | static void unlock_page_lru(struct page *page, int isolated) | |
2379 | { | |
2380 | struct zone *zone = page_zone(page); | |
2381 | ||
2382 | if (isolated) { | |
2383 | struct lruvec *lruvec; | |
2384 | ||
599d0c95 | 2385 | lruvec = mem_cgroup_page_lruvec(page, zone->zone_pgdat); |
0a31bc97 JW |
2386 | VM_BUG_ON_PAGE(PageLRU(page), page); |
2387 | SetPageLRU(page); | |
2388 | add_page_to_lru_list(page, lruvec, page_lru(page)); | |
2389 | } | |
a52633d8 | 2390 | spin_unlock_irq(zone_lru_lock(zone)); |
0a31bc97 JW |
2391 | } |
2392 | ||
00501b53 | 2393 | static void commit_charge(struct page *page, struct mem_cgroup *memcg, |
6abb5a86 | 2394 | bool lrucare) |
7a81b88c | 2395 | { |
0a31bc97 | 2396 | int isolated; |
9ce70c02 | 2397 | |
1306a85a | 2398 | VM_BUG_ON_PAGE(page->mem_cgroup, page); |
9ce70c02 HD |
2399 | |
2400 | /* | |
2401 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page | |
2402 | * may already be on some other mem_cgroup's LRU. Take care of it. | |
2403 | */ | |
0a31bc97 JW |
2404 | if (lrucare) |
2405 | lock_page_lru(page, &isolated); | |
9ce70c02 | 2406 | |
0a31bc97 JW |
2407 | /* |
2408 | * Nobody should be changing or seriously looking at | |
1306a85a | 2409 | * page->mem_cgroup at this point: |
0a31bc97 JW |
2410 | * |
2411 | * - the page is uncharged | |
2412 | * | |
2413 | * - the page is off-LRU | |
2414 | * | |
2415 | * - an anonymous fault has exclusive page access, except for | |
2416 | * a locked page table | |
2417 | * | |
2418 | * - a page cache insertion, a swapin fault, or a migration | |
2419 | * have the page locked | |
2420 | */ | |
1306a85a | 2421 | page->mem_cgroup = memcg; |
9ce70c02 | 2422 | |
0a31bc97 JW |
2423 | if (lrucare) |
2424 | unlock_page_lru(page, isolated); | |
7a81b88c | 2425 | } |
66e1707b | 2426 | |
84c07d11 | 2427 | #ifdef CONFIG_MEMCG_KMEM |
f3bb3043 | 2428 | static int memcg_alloc_cache_id(void) |
55007d84 | 2429 | { |
f3bb3043 VD |
2430 | int id, size; |
2431 | int err; | |
2432 | ||
dbcf73e2 | 2433 | id = ida_simple_get(&memcg_cache_ida, |
f3bb3043 VD |
2434 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); |
2435 | if (id < 0) | |
2436 | return id; | |
55007d84 | 2437 | |
dbcf73e2 | 2438 | if (id < memcg_nr_cache_ids) |
f3bb3043 VD |
2439 | return id; |
2440 | ||
2441 | /* | |
2442 | * There's no space for the new id in memcg_caches arrays, | |
2443 | * so we have to grow them. | |
2444 | */ | |
05257a1a | 2445 | down_write(&memcg_cache_ids_sem); |
f3bb3043 VD |
2446 | |
2447 | size = 2 * (id + 1); | |
55007d84 GC |
2448 | if (size < MEMCG_CACHES_MIN_SIZE) |
2449 | size = MEMCG_CACHES_MIN_SIZE; | |
2450 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
2451 | size = MEMCG_CACHES_MAX_SIZE; | |
2452 | ||
f3bb3043 | 2453 | err = memcg_update_all_caches(size); |
60d3fd32 VD |
2454 | if (!err) |
2455 | err = memcg_update_all_list_lrus(size); | |
05257a1a VD |
2456 | if (!err) |
2457 | memcg_nr_cache_ids = size; | |
2458 | ||
2459 | up_write(&memcg_cache_ids_sem); | |
2460 | ||
f3bb3043 | 2461 | if (err) { |
dbcf73e2 | 2462 | ida_simple_remove(&memcg_cache_ida, id); |
f3bb3043 VD |
2463 | return err; |
2464 | } | |
2465 | return id; | |
2466 | } | |
2467 | ||
2468 | static void memcg_free_cache_id(int id) | |
2469 | { | |
dbcf73e2 | 2470 | ida_simple_remove(&memcg_cache_ida, id); |
55007d84 GC |
2471 | } |
2472 | ||
d5b3cf71 | 2473 | struct memcg_kmem_cache_create_work { |
5722d094 VD |
2474 | struct mem_cgroup *memcg; |
2475 | struct kmem_cache *cachep; | |
2476 | struct work_struct work; | |
2477 | }; | |
2478 | ||
d5b3cf71 | 2479 | static void memcg_kmem_cache_create_func(struct work_struct *w) |
d7f25f8a | 2480 | { |
d5b3cf71 VD |
2481 | struct memcg_kmem_cache_create_work *cw = |
2482 | container_of(w, struct memcg_kmem_cache_create_work, work); | |
5722d094 VD |
2483 | struct mem_cgroup *memcg = cw->memcg; |
2484 | struct kmem_cache *cachep = cw->cachep; | |
d7f25f8a | 2485 | |
d5b3cf71 | 2486 | memcg_create_kmem_cache(memcg, cachep); |
bd673145 | 2487 | |
5722d094 | 2488 | css_put(&memcg->css); |
d7f25f8a GC |
2489 | kfree(cw); |
2490 | } | |
2491 | ||
2492 | /* | |
2493 | * Enqueue the creation of a per-memcg kmem_cache. | |
d7f25f8a | 2494 | */ |
85cfb245 | 2495 | static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
d5b3cf71 | 2496 | struct kmem_cache *cachep) |
d7f25f8a | 2497 | { |
d5b3cf71 | 2498 | struct memcg_kmem_cache_create_work *cw; |
d7f25f8a | 2499 | |
c892fd82 | 2500 | cw = kmalloc(sizeof(*cw), GFP_NOWAIT | __GFP_NOWARN); |
8135be5a | 2501 | if (!cw) |
d7f25f8a | 2502 | return; |
8135be5a VD |
2503 | |
2504 | css_get(&memcg->css); | |
d7f25f8a GC |
2505 | |
2506 | cw->memcg = memcg; | |
2507 | cw->cachep = cachep; | |
d5b3cf71 | 2508 | INIT_WORK(&cw->work, memcg_kmem_cache_create_func); |
d7f25f8a | 2509 | |
17cc4dfe | 2510 | queue_work(memcg_kmem_cache_wq, &cw->work); |
d7f25f8a GC |
2511 | } |
2512 | ||
45264778 VD |
2513 | static inline bool memcg_kmem_bypass(void) |
2514 | { | |
2515 | if (in_interrupt() || !current->mm || (current->flags & PF_KTHREAD)) | |
2516 | return true; | |
2517 | return false; | |
2518 | } | |
2519 | ||
2520 | /** | |
2521 | * memcg_kmem_get_cache: select the correct per-memcg cache for allocation | |
2522 | * @cachep: the original global kmem cache | |
2523 | * | |
d7f25f8a GC |
2524 | * Return the kmem_cache we're supposed to use for a slab allocation. |
2525 | * We try to use the current memcg's version of the cache. | |
2526 | * | |
45264778 VD |
2527 | * If the cache does not exist yet, if we are the first user of it, we |
2528 | * create it asynchronously in a workqueue and let the current allocation | |
2529 | * go through with the original cache. | |
d7f25f8a | 2530 | * |
45264778 VD |
2531 | * This function takes a reference to the cache it returns to assure it |
2532 | * won't get destroyed while we are working with it. Once the caller is | |
2533 | * done with it, memcg_kmem_put_cache() must be called to release the | |
2534 | * reference. | |
d7f25f8a | 2535 | */ |
45264778 | 2536 | struct kmem_cache *memcg_kmem_get_cache(struct kmem_cache *cachep) |
d7f25f8a GC |
2537 | { |
2538 | struct mem_cgroup *memcg; | |
959c8963 | 2539 | struct kmem_cache *memcg_cachep; |
2a4db7eb | 2540 | int kmemcg_id; |
d7f25f8a | 2541 | |
f7ce3190 | 2542 | VM_BUG_ON(!is_root_cache(cachep)); |
d7f25f8a | 2543 | |
45264778 | 2544 | if (memcg_kmem_bypass()) |
230e9fc2 VD |
2545 | return cachep; |
2546 | ||
d46eb14b | 2547 | memcg = get_mem_cgroup_from_current(); |
4db0c3c2 | 2548 | kmemcg_id = READ_ONCE(memcg->kmemcg_id); |
2a4db7eb | 2549 | if (kmemcg_id < 0) |
ca0dde97 | 2550 | goto out; |
d7f25f8a | 2551 | |
2a4db7eb | 2552 | memcg_cachep = cache_from_memcg_idx(cachep, kmemcg_id); |
8135be5a VD |
2553 | if (likely(memcg_cachep)) |
2554 | return memcg_cachep; | |
ca0dde97 LZ |
2555 | |
2556 | /* | |
2557 | * If we are in a safe context (can wait, and not in interrupt | |
2558 | * context), we could be be predictable and return right away. | |
2559 | * This would guarantee that the allocation being performed | |
2560 | * already belongs in the new cache. | |
2561 | * | |
2562 | * However, there are some clashes that can arrive from locking. | |
2563 | * For instance, because we acquire the slab_mutex while doing | |
776ed0f0 VD |
2564 | * memcg_create_kmem_cache, this means no further allocation |
2565 | * could happen with the slab_mutex held. So it's better to | |
2566 | * defer everything. | |
ca0dde97 | 2567 | */ |
d5b3cf71 | 2568 | memcg_schedule_kmem_cache_create(memcg, cachep); |
ca0dde97 | 2569 | out: |
8135be5a | 2570 | css_put(&memcg->css); |
ca0dde97 | 2571 | return cachep; |
d7f25f8a | 2572 | } |
d7f25f8a | 2573 | |
45264778 VD |
2574 | /** |
2575 | * memcg_kmem_put_cache: drop reference taken by memcg_kmem_get_cache | |
2576 | * @cachep: the cache returned by memcg_kmem_get_cache | |
2577 | */ | |
2578 | void memcg_kmem_put_cache(struct kmem_cache *cachep) | |
8135be5a VD |
2579 | { |
2580 | if (!is_root_cache(cachep)) | |
f7ce3190 | 2581 | css_put(&cachep->memcg_params.memcg->css); |
8135be5a VD |
2582 | } |
2583 | ||
45264778 | 2584 | /** |
60cd4bcd | 2585 | * __memcg_kmem_charge_memcg: charge a kmem page |
45264778 VD |
2586 | * @page: page to charge |
2587 | * @gfp: reclaim mode | |
2588 | * @order: allocation order | |
2589 | * @memcg: memory cgroup to charge | |
2590 | * | |
2591 | * Returns 0 on success, an error code on failure. | |
2592 | */ | |
60cd4bcd | 2593 | int __memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order, |
45264778 | 2594 | struct mem_cgroup *memcg) |
7ae1e1d0 | 2595 | { |
f3ccb2c4 VD |
2596 | unsigned int nr_pages = 1 << order; |
2597 | struct page_counter *counter; | |
7ae1e1d0 GC |
2598 | int ret; |
2599 | ||
f3ccb2c4 | 2600 | ret = try_charge(memcg, gfp, nr_pages); |
52c29b04 | 2601 | if (ret) |
f3ccb2c4 | 2602 | return ret; |
52c29b04 JW |
2603 | |
2604 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && | |
2605 | !page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) { | |
2606 | cancel_charge(memcg, nr_pages); | |
2607 | return -ENOMEM; | |
7ae1e1d0 GC |
2608 | } |
2609 | ||
f3ccb2c4 | 2610 | page->mem_cgroup = memcg; |
7ae1e1d0 | 2611 | |
f3ccb2c4 | 2612 | return 0; |
7ae1e1d0 GC |
2613 | } |
2614 | ||
45264778 | 2615 | /** |
60cd4bcd | 2616 | * __memcg_kmem_charge: charge a kmem page to the current memory cgroup |
45264778 VD |
2617 | * @page: page to charge |
2618 | * @gfp: reclaim mode | |
2619 | * @order: allocation order | |
2620 | * | |
2621 | * Returns 0 on success, an error code on failure. | |
2622 | */ | |
60cd4bcd | 2623 | int __memcg_kmem_charge(struct page *page, gfp_t gfp, int order) |
7ae1e1d0 | 2624 | { |
f3ccb2c4 | 2625 | struct mem_cgroup *memcg; |
fcff7d7e | 2626 | int ret = 0; |
7ae1e1d0 | 2627 | |
60cd4bcd | 2628 | if (memcg_kmem_bypass()) |
45264778 VD |
2629 | return 0; |
2630 | ||
d46eb14b | 2631 | memcg = get_mem_cgroup_from_current(); |
c4159a75 | 2632 | if (!mem_cgroup_is_root(memcg)) { |
60cd4bcd | 2633 | ret = __memcg_kmem_charge_memcg(page, gfp, order, memcg); |
c4159a75 VD |
2634 | if (!ret) |
2635 | __SetPageKmemcg(page); | |
2636 | } | |
7ae1e1d0 | 2637 | css_put(&memcg->css); |
d05e83a6 | 2638 | return ret; |
7ae1e1d0 | 2639 | } |
45264778 | 2640 | /** |
60cd4bcd | 2641 | * __memcg_kmem_uncharge: uncharge a kmem page |
45264778 VD |
2642 | * @page: page to uncharge |
2643 | * @order: allocation order | |
2644 | */ | |
60cd4bcd | 2645 | void __memcg_kmem_uncharge(struct page *page, int order) |
7ae1e1d0 | 2646 | { |
1306a85a | 2647 | struct mem_cgroup *memcg = page->mem_cgroup; |
f3ccb2c4 | 2648 | unsigned int nr_pages = 1 << order; |
7ae1e1d0 | 2649 | |
7ae1e1d0 GC |
2650 | if (!memcg) |
2651 | return; | |
2652 | ||
309381fe | 2653 | VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page); |
29833315 | 2654 | |
52c29b04 JW |
2655 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
2656 | page_counter_uncharge(&memcg->kmem, nr_pages); | |
2657 | ||
f3ccb2c4 | 2658 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 2659 | if (do_memsw_account()) |
f3ccb2c4 | 2660 | page_counter_uncharge(&memcg->memsw, nr_pages); |
60d3fd32 | 2661 | |
1306a85a | 2662 | page->mem_cgroup = NULL; |
c4159a75 VD |
2663 | |
2664 | /* slab pages do not have PageKmemcg flag set */ | |
2665 | if (PageKmemcg(page)) | |
2666 | __ClearPageKmemcg(page); | |
2667 | ||
f3ccb2c4 | 2668 | css_put_many(&memcg->css, nr_pages); |
60d3fd32 | 2669 | } |
84c07d11 | 2670 | #endif /* CONFIG_MEMCG_KMEM */ |
7ae1e1d0 | 2671 | |
ca3e0214 KH |
2672 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
2673 | ||
ca3e0214 KH |
2674 | /* |
2675 | * Because tail pages are not marked as "used", set it. We're under | |
a52633d8 | 2676 | * zone_lru_lock and migration entries setup in all page mappings. |
ca3e0214 | 2677 | */ |
e94c8a9c | 2678 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 | 2679 | { |
e94c8a9c | 2680 | int i; |
ca3e0214 | 2681 | |
3d37c4a9 KH |
2682 | if (mem_cgroup_disabled()) |
2683 | return; | |
b070e65c | 2684 | |
29833315 | 2685 | for (i = 1; i < HPAGE_PMD_NR; i++) |
1306a85a | 2686 | head[i].mem_cgroup = head->mem_cgroup; |
b9982f8d | 2687 | |
c9019e9b | 2688 | __mod_memcg_state(head->mem_cgroup, MEMCG_RSS_HUGE, -HPAGE_PMD_NR); |
ca3e0214 | 2689 | } |
12d27107 | 2690 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 2691 | |
c255a458 | 2692 | #ifdef CONFIG_MEMCG_SWAP |
02491447 DN |
2693 | /** |
2694 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
2695 | * @entry: swap entry to be moved | |
2696 | * @from: mem_cgroup which the entry is moved from | |
2697 | * @to: mem_cgroup which the entry is moved to | |
2698 | * | |
2699 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
2700 | * as the mem_cgroup's id of @from. | |
2701 | * | |
2702 | * Returns 0 on success, -EINVAL on failure. | |
2703 | * | |
3e32cb2e | 2704 | * The caller must have charged to @to, IOW, called page_counter_charge() about |
02491447 DN |
2705 | * both res and memsw, and called css_get(). |
2706 | */ | |
2707 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 2708 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
2709 | { |
2710 | unsigned short old_id, new_id; | |
2711 | ||
34c00c31 LZ |
2712 | old_id = mem_cgroup_id(from); |
2713 | new_id = mem_cgroup_id(to); | |
02491447 DN |
2714 | |
2715 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
c9019e9b JW |
2716 | mod_memcg_state(from, MEMCG_SWAP, -1); |
2717 | mod_memcg_state(to, MEMCG_SWAP, 1); | |
02491447 DN |
2718 | return 0; |
2719 | } | |
2720 | return -EINVAL; | |
2721 | } | |
2722 | #else | |
2723 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 2724 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
2725 | { |
2726 | return -EINVAL; | |
2727 | } | |
8c7c6e34 | 2728 | #endif |
d13d1443 | 2729 | |
bbec2e15 | 2730 | static DEFINE_MUTEX(memcg_max_mutex); |
f212ad7c | 2731 | |
bbec2e15 RG |
2732 | static int mem_cgroup_resize_max(struct mem_cgroup *memcg, |
2733 | unsigned long max, bool memsw) | |
628f4235 | 2734 | { |
3e32cb2e | 2735 | bool enlarge = false; |
bb4a7ea2 | 2736 | bool drained = false; |
3e32cb2e | 2737 | int ret; |
c054a78c YZ |
2738 | bool limits_invariant; |
2739 | struct page_counter *counter = memsw ? &memcg->memsw : &memcg->memory; | |
81d39c20 | 2740 | |
3e32cb2e | 2741 | do { |
628f4235 KH |
2742 | if (signal_pending(current)) { |
2743 | ret = -EINTR; | |
2744 | break; | |
2745 | } | |
3e32cb2e | 2746 | |
bbec2e15 | 2747 | mutex_lock(&memcg_max_mutex); |
c054a78c YZ |
2748 | /* |
2749 | * Make sure that the new limit (memsw or memory limit) doesn't | |
bbec2e15 | 2750 | * break our basic invariant rule memory.max <= memsw.max. |
c054a78c | 2751 | */ |
bbec2e15 RG |
2752 | limits_invariant = memsw ? max >= memcg->memory.max : |
2753 | max <= memcg->memsw.max; | |
c054a78c | 2754 | if (!limits_invariant) { |
bbec2e15 | 2755 | mutex_unlock(&memcg_max_mutex); |
8c7c6e34 | 2756 | ret = -EINVAL; |
8c7c6e34 KH |
2757 | break; |
2758 | } | |
bbec2e15 | 2759 | if (max > counter->max) |
3e32cb2e | 2760 | enlarge = true; |
bbec2e15 RG |
2761 | ret = page_counter_set_max(counter, max); |
2762 | mutex_unlock(&memcg_max_mutex); | |
8c7c6e34 KH |
2763 | |
2764 | if (!ret) | |
2765 | break; | |
2766 | ||
bb4a7ea2 SB |
2767 | if (!drained) { |
2768 | drain_all_stock(memcg); | |
2769 | drained = true; | |
2770 | continue; | |
2771 | } | |
2772 | ||
1ab5c056 AR |
2773 | if (!try_to_free_mem_cgroup_pages(memcg, 1, |
2774 | GFP_KERNEL, !memsw)) { | |
2775 | ret = -EBUSY; | |
2776 | break; | |
2777 | } | |
2778 | } while (true); | |
3e32cb2e | 2779 | |
3c11ecf4 KH |
2780 | if (!ret && enlarge) |
2781 | memcg_oom_recover(memcg); | |
3e32cb2e | 2782 | |
628f4235 KH |
2783 | return ret; |
2784 | } | |
2785 | ||
ef8f2327 | 2786 | unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order, |
0608f43d AM |
2787 | gfp_t gfp_mask, |
2788 | unsigned long *total_scanned) | |
2789 | { | |
2790 | unsigned long nr_reclaimed = 0; | |
ef8f2327 | 2791 | struct mem_cgroup_per_node *mz, *next_mz = NULL; |
0608f43d AM |
2792 | unsigned long reclaimed; |
2793 | int loop = 0; | |
ef8f2327 | 2794 | struct mem_cgroup_tree_per_node *mctz; |
3e32cb2e | 2795 | unsigned long excess; |
0608f43d AM |
2796 | unsigned long nr_scanned; |
2797 | ||
2798 | if (order > 0) | |
2799 | return 0; | |
2800 | ||
ef8f2327 | 2801 | mctz = soft_limit_tree_node(pgdat->node_id); |
d6507ff5 MH |
2802 | |
2803 | /* | |
2804 | * Do not even bother to check the largest node if the root | |
2805 | * is empty. Do it lockless to prevent lock bouncing. Races | |
2806 | * are acceptable as soft limit is best effort anyway. | |
2807 | */ | |
bfc7228b | 2808 | if (!mctz || RB_EMPTY_ROOT(&mctz->rb_root)) |
d6507ff5 MH |
2809 | return 0; |
2810 | ||
0608f43d AM |
2811 | /* |
2812 | * This loop can run a while, specially if mem_cgroup's continuously | |
2813 | * keep exceeding their soft limit and putting the system under | |
2814 | * pressure | |
2815 | */ | |
2816 | do { | |
2817 | if (next_mz) | |
2818 | mz = next_mz; | |
2819 | else | |
2820 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
2821 | if (!mz) | |
2822 | break; | |
2823 | ||
2824 | nr_scanned = 0; | |
ef8f2327 | 2825 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, pgdat, |
0608f43d AM |
2826 | gfp_mask, &nr_scanned); |
2827 | nr_reclaimed += reclaimed; | |
2828 | *total_scanned += nr_scanned; | |
0a31bc97 | 2829 | spin_lock_irq(&mctz->lock); |
bc2f2e7f | 2830 | __mem_cgroup_remove_exceeded(mz, mctz); |
0608f43d AM |
2831 | |
2832 | /* | |
2833 | * If we failed to reclaim anything from this memory cgroup | |
2834 | * it is time to move on to the next cgroup | |
2835 | */ | |
2836 | next_mz = NULL; | |
bc2f2e7f VD |
2837 | if (!reclaimed) |
2838 | next_mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
2839 | ||
3e32cb2e | 2840 | excess = soft_limit_excess(mz->memcg); |
0608f43d AM |
2841 | /* |
2842 | * One school of thought says that we should not add | |
2843 | * back the node to the tree if reclaim returns 0. | |
2844 | * But our reclaim could return 0, simply because due | |
2845 | * to priority we are exposing a smaller subset of | |
2846 | * memory to reclaim from. Consider this as a longer | |
2847 | * term TODO. | |
2848 | */ | |
2849 | /* If excess == 0, no tree ops */ | |
cf2c8127 | 2850 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 2851 | spin_unlock_irq(&mctz->lock); |
0608f43d AM |
2852 | css_put(&mz->memcg->css); |
2853 | loop++; | |
2854 | /* | |
2855 | * Could not reclaim anything and there are no more | |
2856 | * mem cgroups to try or we seem to be looping without | |
2857 | * reclaiming anything. | |
2858 | */ | |
2859 | if (!nr_reclaimed && | |
2860 | (next_mz == NULL || | |
2861 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
2862 | break; | |
2863 | } while (!nr_reclaimed); | |
2864 | if (next_mz) | |
2865 | css_put(&next_mz->memcg->css); | |
2866 | return nr_reclaimed; | |
2867 | } | |
2868 | ||
ea280e7b TH |
2869 | /* |
2870 | * Test whether @memcg has children, dead or alive. Note that this | |
2871 | * function doesn't care whether @memcg has use_hierarchy enabled and | |
2872 | * returns %true if there are child csses according to the cgroup | |
2873 | * hierarchy. Testing use_hierarchy is the caller's responsiblity. | |
2874 | */ | |
b5f99b53 GC |
2875 | static inline bool memcg_has_children(struct mem_cgroup *memcg) |
2876 | { | |
ea280e7b TH |
2877 | bool ret; |
2878 | ||
ea280e7b TH |
2879 | rcu_read_lock(); |
2880 | ret = css_next_child(NULL, &memcg->css); | |
2881 | rcu_read_unlock(); | |
2882 | return ret; | |
b5f99b53 GC |
2883 | } |
2884 | ||
c26251f9 | 2885 | /* |
51038171 | 2886 | * Reclaims as many pages from the given memcg as possible. |
c26251f9 MH |
2887 | * |
2888 | * Caller is responsible for holding css reference for memcg. | |
2889 | */ | |
2890 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
2891 | { | |
2892 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c26251f9 | 2893 | |
c1e862c1 KH |
2894 | /* we call try-to-free pages for make this cgroup empty */ |
2895 | lru_add_drain_all(); | |
d12c60f6 JS |
2896 | |
2897 | drain_all_stock(memcg); | |
2898 | ||
f817ed48 | 2899 | /* try to free all pages in this cgroup */ |
3e32cb2e | 2900 | while (nr_retries && page_counter_read(&memcg->memory)) { |
f817ed48 | 2901 | int progress; |
c1e862c1 | 2902 | |
c26251f9 MH |
2903 | if (signal_pending(current)) |
2904 | return -EINTR; | |
2905 | ||
b70a2a21 JW |
2906 | progress = try_to_free_mem_cgroup_pages(memcg, 1, |
2907 | GFP_KERNEL, true); | |
c1e862c1 | 2908 | if (!progress) { |
f817ed48 | 2909 | nr_retries--; |
c1e862c1 | 2910 | /* maybe some writeback is necessary */ |
8aa7e847 | 2911 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 2912 | } |
f817ed48 KH |
2913 | |
2914 | } | |
ab5196c2 MH |
2915 | |
2916 | return 0; | |
cc847582 KH |
2917 | } |
2918 | ||
6770c64e TH |
2919 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
2920 | char *buf, size_t nbytes, | |
2921 | loff_t off) | |
c1e862c1 | 2922 | { |
6770c64e | 2923 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
c26251f9 | 2924 | |
d8423011 MH |
2925 | if (mem_cgroup_is_root(memcg)) |
2926 | return -EINVAL; | |
6770c64e | 2927 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
c1e862c1 KH |
2928 | } |
2929 | ||
182446d0 TH |
2930 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
2931 | struct cftype *cft) | |
18f59ea7 | 2932 | { |
182446d0 | 2933 | return mem_cgroup_from_css(css)->use_hierarchy; |
18f59ea7 BS |
2934 | } |
2935 | ||
182446d0 TH |
2936 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
2937 | struct cftype *cft, u64 val) | |
18f59ea7 BS |
2938 | { |
2939 | int retval = 0; | |
182446d0 | 2940 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 2941 | struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent); |
18f59ea7 | 2942 | |
567fb435 | 2943 | if (memcg->use_hierarchy == val) |
0b8f73e1 | 2944 | return 0; |
567fb435 | 2945 | |
18f59ea7 | 2946 | /* |
af901ca1 | 2947 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
2948 | * in the child subtrees. If it is unset, then the change can |
2949 | * occur, provided the current cgroup has no children. | |
2950 | * | |
2951 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
2952 | * set if there are no children. | |
2953 | */ | |
c0ff4b85 | 2954 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 | 2955 | (val == 1 || val == 0)) { |
ea280e7b | 2956 | if (!memcg_has_children(memcg)) |
c0ff4b85 | 2957 | memcg->use_hierarchy = val; |
18f59ea7 BS |
2958 | else |
2959 | retval = -EBUSY; | |
2960 | } else | |
2961 | retval = -EINVAL; | |
567fb435 | 2962 | |
18f59ea7 BS |
2963 | return retval; |
2964 | } | |
2965 | ||
8de7ecc6 SB |
2966 | struct accumulated_stats { |
2967 | unsigned long stat[MEMCG_NR_STAT]; | |
2968 | unsigned long events[NR_VM_EVENT_ITEMS]; | |
2969 | unsigned long lru_pages[NR_LRU_LISTS]; | |
2970 | const unsigned int *stats_array; | |
2971 | const unsigned int *events_array; | |
2972 | int stats_size; | |
2973 | int events_size; | |
2974 | }; | |
ce00a967 | 2975 | |
8de7ecc6 SB |
2976 | static void accumulate_memcg_tree(struct mem_cgroup *memcg, |
2977 | struct accumulated_stats *acc) | |
587d9f72 | 2978 | { |
8de7ecc6 | 2979 | struct mem_cgroup *mi; |
72b54e73 | 2980 | int i; |
587d9f72 | 2981 | |
8de7ecc6 SB |
2982 | for_each_mem_cgroup_tree(mi, memcg) { |
2983 | for (i = 0; i < acc->stats_size; i++) | |
2984 | acc->stat[i] += memcg_page_state(mi, | |
2985 | acc->stats_array ? acc->stats_array[i] : i); | |
587d9f72 | 2986 | |
8de7ecc6 SB |
2987 | for (i = 0; i < acc->events_size; i++) |
2988 | acc->events[i] += memcg_sum_events(mi, | |
2989 | acc->events_array ? acc->events_array[i] : i); | |
2990 | ||
2991 | for (i = 0; i < NR_LRU_LISTS; i++) | |
2992 | acc->lru_pages[i] += | |
2993 | mem_cgroup_nr_lru_pages(mi, BIT(i)); | |
72b54e73 | 2994 | } |
587d9f72 JW |
2995 | } |
2996 | ||
6f646156 | 2997 | static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
ce00a967 | 2998 | { |
72b54e73 | 2999 | unsigned long val = 0; |
ce00a967 | 3000 | |
3e32cb2e | 3001 | if (mem_cgroup_is_root(memcg)) { |
72b54e73 VD |
3002 | struct mem_cgroup *iter; |
3003 | ||
3004 | for_each_mem_cgroup_tree(iter, memcg) { | |
ccda7f43 JW |
3005 | val += memcg_page_state(iter, MEMCG_CACHE); |
3006 | val += memcg_page_state(iter, MEMCG_RSS); | |
72b54e73 | 3007 | if (swap) |
ccda7f43 | 3008 | val += memcg_page_state(iter, MEMCG_SWAP); |
72b54e73 | 3009 | } |
3e32cb2e | 3010 | } else { |
ce00a967 | 3011 | if (!swap) |
3e32cb2e | 3012 | val = page_counter_read(&memcg->memory); |
ce00a967 | 3013 | else |
3e32cb2e | 3014 | val = page_counter_read(&memcg->memsw); |
ce00a967 | 3015 | } |
c12176d3 | 3016 | return val; |
ce00a967 JW |
3017 | } |
3018 | ||
3e32cb2e JW |
3019 | enum { |
3020 | RES_USAGE, | |
3021 | RES_LIMIT, | |
3022 | RES_MAX_USAGE, | |
3023 | RES_FAILCNT, | |
3024 | RES_SOFT_LIMIT, | |
3025 | }; | |
ce00a967 | 3026 | |
791badbd | 3027 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
05b84301 | 3028 | struct cftype *cft) |
8cdea7c0 | 3029 | { |
182446d0 | 3030 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3e32cb2e | 3031 | struct page_counter *counter; |
af36f906 | 3032 | |
3e32cb2e | 3033 | switch (MEMFILE_TYPE(cft->private)) { |
8c7c6e34 | 3034 | case _MEM: |
3e32cb2e JW |
3035 | counter = &memcg->memory; |
3036 | break; | |
8c7c6e34 | 3037 | case _MEMSWAP: |
3e32cb2e JW |
3038 | counter = &memcg->memsw; |
3039 | break; | |
510fc4e1 | 3040 | case _KMEM: |
3e32cb2e | 3041 | counter = &memcg->kmem; |
510fc4e1 | 3042 | break; |
d55f90bf | 3043 | case _TCP: |
0db15298 | 3044 | counter = &memcg->tcpmem; |
d55f90bf | 3045 | break; |
8c7c6e34 KH |
3046 | default: |
3047 | BUG(); | |
8c7c6e34 | 3048 | } |
3e32cb2e JW |
3049 | |
3050 | switch (MEMFILE_ATTR(cft->private)) { | |
3051 | case RES_USAGE: | |
3052 | if (counter == &memcg->memory) | |
c12176d3 | 3053 | return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE; |
3e32cb2e | 3054 | if (counter == &memcg->memsw) |
c12176d3 | 3055 | return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE; |
3e32cb2e JW |
3056 | return (u64)page_counter_read(counter) * PAGE_SIZE; |
3057 | case RES_LIMIT: | |
bbec2e15 | 3058 | return (u64)counter->max * PAGE_SIZE; |
3e32cb2e JW |
3059 | case RES_MAX_USAGE: |
3060 | return (u64)counter->watermark * PAGE_SIZE; | |
3061 | case RES_FAILCNT: | |
3062 | return counter->failcnt; | |
3063 | case RES_SOFT_LIMIT: | |
3064 | return (u64)memcg->soft_limit * PAGE_SIZE; | |
3065 | default: | |
3066 | BUG(); | |
3067 | } | |
8cdea7c0 | 3068 | } |
510fc4e1 | 3069 | |
84c07d11 | 3070 | #ifdef CONFIG_MEMCG_KMEM |
567e9ab2 | 3071 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
d6441637 | 3072 | { |
d6441637 VD |
3073 | int memcg_id; |
3074 | ||
b313aeee VD |
3075 | if (cgroup_memory_nokmem) |
3076 | return 0; | |
3077 | ||
2a4db7eb | 3078 | BUG_ON(memcg->kmemcg_id >= 0); |
567e9ab2 | 3079 | BUG_ON(memcg->kmem_state); |
d6441637 | 3080 | |
f3bb3043 | 3081 | memcg_id = memcg_alloc_cache_id(); |
0b8f73e1 JW |
3082 | if (memcg_id < 0) |
3083 | return memcg_id; | |
d6441637 | 3084 | |
ef12947c | 3085 | static_branch_inc(&memcg_kmem_enabled_key); |
d6441637 | 3086 | /* |
567e9ab2 | 3087 | * A memory cgroup is considered kmem-online as soon as it gets |
900a38f0 | 3088 | * kmemcg_id. Setting the id after enabling static branching will |
d6441637 VD |
3089 | * guarantee no one starts accounting before all call sites are |
3090 | * patched. | |
3091 | */ | |
900a38f0 | 3092 | memcg->kmemcg_id = memcg_id; |
567e9ab2 | 3093 | memcg->kmem_state = KMEM_ONLINE; |
bc2791f8 | 3094 | INIT_LIST_HEAD(&memcg->kmem_caches); |
0b8f73e1 JW |
3095 | |
3096 | return 0; | |
d6441637 VD |
3097 | } |
3098 | ||
8e0a8912 JW |
3099 | static void memcg_offline_kmem(struct mem_cgroup *memcg) |
3100 | { | |
3101 | struct cgroup_subsys_state *css; | |
3102 | struct mem_cgroup *parent, *child; | |
3103 | int kmemcg_id; | |
3104 | ||
3105 | if (memcg->kmem_state != KMEM_ONLINE) | |
3106 | return; | |
3107 | /* | |
3108 | * Clear the online state before clearing memcg_caches array | |
3109 | * entries. The slab_mutex in memcg_deactivate_kmem_caches() | |
3110 | * guarantees that no cache will be created for this cgroup | |
3111 | * after we are done (see memcg_create_kmem_cache()). | |
3112 | */ | |
3113 | memcg->kmem_state = KMEM_ALLOCATED; | |
3114 | ||
3115 | memcg_deactivate_kmem_caches(memcg); | |
3116 | ||
3117 | kmemcg_id = memcg->kmemcg_id; | |
3118 | BUG_ON(kmemcg_id < 0); | |
3119 | ||
3120 | parent = parent_mem_cgroup(memcg); | |
3121 | if (!parent) | |
3122 | parent = root_mem_cgroup; | |
3123 | ||
3124 | /* | |
3125 | * Change kmemcg_id of this cgroup and all its descendants to the | |
3126 | * parent's id, and then move all entries from this cgroup's list_lrus | |
3127 | * to ones of the parent. After we have finished, all list_lrus | |
3128 | * corresponding to this cgroup are guaranteed to remain empty. The | |
3129 | * ordering is imposed by list_lru_node->lock taken by | |
3130 | * memcg_drain_all_list_lrus(). | |
3131 | */ | |
3a06bb78 | 3132 | rcu_read_lock(); /* can be called from css_free w/o cgroup_mutex */ |
8e0a8912 JW |
3133 | css_for_each_descendant_pre(css, &memcg->css) { |
3134 | child = mem_cgroup_from_css(css); | |
3135 | BUG_ON(child->kmemcg_id != kmemcg_id); | |
3136 | child->kmemcg_id = parent->kmemcg_id; | |
3137 | if (!memcg->use_hierarchy) | |
3138 | break; | |
3139 | } | |
3a06bb78 TH |
3140 | rcu_read_unlock(); |
3141 | ||
9bec5c35 | 3142 | memcg_drain_all_list_lrus(kmemcg_id, parent); |
8e0a8912 JW |
3143 | |
3144 | memcg_free_cache_id(kmemcg_id); | |
3145 | } | |
3146 | ||
3147 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
3148 | { | |
0b8f73e1 JW |
3149 | /* css_alloc() failed, offlining didn't happen */ |
3150 | if (unlikely(memcg->kmem_state == KMEM_ONLINE)) | |
3151 | memcg_offline_kmem(memcg); | |
3152 | ||
8e0a8912 JW |
3153 | if (memcg->kmem_state == KMEM_ALLOCATED) { |
3154 | memcg_destroy_kmem_caches(memcg); | |
3155 | static_branch_dec(&memcg_kmem_enabled_key); | |
3156 | WARN_ON(page_counter_read(&memcg->kmem)); | |
3157 | } | |
8e0a8912 | 3158 | } |
d6441637 | 3159 | #else |
0b8f73e1 | 3160 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
127424c8 JW |
3161 | { |
3162 | return 0; | |
3163 | } | |
3164 | static void memcg_offline_kmem(struct mem_cgroup *memcg) | |
3165 | { | |
3166 | } | |
3167 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
3168 | { | |
3169 | } | |
84c07d11 | 3170 | #endif /* CONFIG_MEMCG_KMEM */ |
127424c8 | 3171 | |
bbec2e15 RG |
3172 | static int memcg_update_kmem_max(struct mem_cgroup *memcg, |
3173 | unsigned long max) | |
d6441637 | 3174 | { |
b313aeee | 3175 | int ret; |
127424c8 | 3176 | |
bbec2e15 RG |
3177 | mutex_lock(&memcg_max_mutex); |
3178 | ret = page_counter_set_max(&memcg->kmem, max); | |
3179 | mutex_unlock(&memcg_max_mutex); | |
127424c8 | 3180 | return ret; |
d6441637 | 3181 | } |
510fc4e1 | 3182 | |
bbec2e15 | 3183 | static int memcg_update_tcp_max(struct mem_cgroup *memcg, unsigned long max) |
d55f90bf VD |
3184 | { |
3185 | int ret; | |
3186 | ||
bbec2e15 | 3187 | mutex_lock(&memcg_max_mutex); |
d55f90bf | 3188 | |
bbec2e15 | 3189 | ret = page_counter_set_max(&memcg->tcpmem, max); |
d55f90bf VD |
3190 | if (ret) |
3191 | goto out; | |
3192 | ||
0db15298 | 3193 | if (!memcg->tcpmem_active) { |
d55f90bf VD |
3194 | /* |
3195 | * The active flag needs to be written after the static_key | |
3196 | * update. This is what guarantees that the socket activation | |
2d758073 JW |
3197 | * function is the last one to run. See mem_cgroup_sk_alloc() |
3198 | * for details, and note that we don't mark any socket as | |
3199 | * belonging to this memcg until that flag is up. | |
d55f90bf VD |
3200 | * |
3201 | * We need to do this, because static_keys will span multiple | |
3202 | * sites, but we can't control their order. If we mark a socket | |
3203 | * as accounted, but the accounting functions are not patched in | |
3204 | * yet, we'll lose accounting. | |
3205 | * | |
2d758073 | 3206 | * We never race with the readers in mem_cgroup_sk_alloc(), |
d55f90bf VD |
3207 | * because when this value change, the code to process it is not |
3208 | * patched in yet. | |
3209 | */ | |
3210 | static_branch_inc(&memcg_sockets_enabled_key); | |
0db15298 | 3211 | memcg->tcpmem_active = true; |
d55f90bf VD |
3212 | } |
3213 | out: | |
bbec2e15 | 3214 | mutex_unlock(&memcg_max_mutex); |
d55f90bf VD |
3215 | return ret; |
3216 | } | |
d55f90bf | 3217 | |
628f4235 KH |
3218 | /* |
3219 | * The user of this function is... | |
3220 | * RES_LIMIT. | |
3221 | */ | |
451af504 TH |
3222 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
3223 | char *buf, size_t nbytes, loff_t off) | |
8cdea7c0 | 3224 | { |
451af504 | 3225 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3226 | unsigned long nr_pages; |
628f4235 KH |
3227 | int ret; |
3228 | ||
451af504 | 3229 | buf = strstrip(buf); |
650c5e56 | 3230 | ret = page_counter_memparse(buf, "-1", &nr_pages); |
3e32cb2e JW |
3231 | if (ret) |
3232 | return ret; | |
af36f906 | 3233 | |
3e32cb2e | 3234 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
628f4235 | 3235 | case RES_LIMIT: |
4b3bde4c BS |
3236 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
3237 | ret = -EINVAL; | |
3238 | break; | |
3239 | } | |
3e32cb2e JW |
3240 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3241 | case _MEM: | |
bbec2e15 | 3242 | ret = mem_cgroup_resize_max(memcg, nr_pages, false); |
8c7c6e34 | 3243 | break; |
3e32cb2e | 3244 | case _MEMSWAP: |
bbec2e15 | 3245 | ret = mem_cgroup_resize_max(memcg, nr_pages, true); |
296c81d8 | 3246 | break; |
3e32cb2e | 3247 | case _KMEM: |
bbec2e15 | 3248 | ret = memcg_update_kmem_max(memcg, nr_pages); |
3e32cb2e | 3249 | break; |
d55f90bf | 3250 | case _TCP: |
bbec2e15 | 3251 | ret = memcg_update_tcp_max(memcg, nr_pages); |
d55f90bf | 3252 | break; |
3e32cb2e | 3253 | } |
296c81d8 | 3254 | break; |
3e32cb2e JW |
3255 | case RES_SOFT_LIMIT: |
3256 | memcg->soft_limit = nr_pages; | |
3257 | ret = 0; | |
628f4235 KH |
3258 | break; |
3259 | } | |
451af504 | 3260 | return ret ?: nbytes; |
8cdea7c0 BS |
3261 | } |
3262 | ||
6770c64e TH |
3263 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
3264 | size_t nbytes, loff_t off) | |
c84872e1 | 3265 | { |
6770c64e | 3266 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3267 | struct page_counter *counter; |
c84872e1 | 3268 | |
3e32cb2e JW |
3269 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3270 | case _MEM: | |
3271 | counter = &memcg->memory; | |
3272 | break; | |
3273 | case _MEMSWAP: | |
3274 | counter = &memcg->memsw; | |
3275 | break; | |
3276 | case _KMEM: | |
3277 | counter = &memcg->kmem; | |
3278 | break; | |
d55f90bf | 3279 | case _TCP: |
0db15298 | 3280 | counter = &memcg->tcpmem; |
d55f90bf | 3281 | break; |
3e32cb2e JW |
3282 | default: |
3283 | BUG(); | |
3284 | } | |
af36f906 | 3285 | |
3e32cb2e | 3286 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
29f2a4da | 3287 | case RES_MAX_USAGE: |
3e32cb2e | 3288 | page_counter_reset_watermark(counter); |
29f2a4da PE |
3289 | break; |
3290 | case RES_FAILCNT: | |
3e32cb2e | 3291 | counter->failcnt = 0; |
29f2a4da | 3292 | break; |
3e32cb2e JW |
3293 | default: |
3294 | BUG(); | |
29f2a4da | 3295 | } |
f64c3f54 | 3296 | |
6770c64e | 3297 | return nbytes; |
c84872e1 PE |
3298 | } |
3299 | ||
182446d0 | 3300 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3301 | struct cftype *cft) |
3302 | { | |
182446d0 | 3303 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
3304 | } |
3305 | ||
02491447 | 3306 | #ifdef CONFIG_MMU |
182446d0 | 3307 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3308 | struct cftype *cft, u64 val) |
3309 | { | |
182446d0 | 3310 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 | 3311 | |
1dfab5ab | 3312 | if (val & ~MOVE_MASK) |
7dc74be0 | 3313 | return -EINVAL; |
ee5e8472 | 3314 | |
7dc74be0 | 3315 | /* |
ee5e8472 GC |
3316 | * No kind of locking is needed in here, because ->can_attach() will |
3317 | * check this value once in the beginning of the process, and then carry | |
3318 | * on with stale data. This means that changes to this value will only | |
3319 | * affect task migrations starting after the change. | |
7dc74be0 | 3320 | */ |
c0ff4b85 | 3321 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
3322 | return 0; |
3323 | } | |
02491447 | 3324 | #else |
182446d0 | 3325 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
3326 | struct cftype *cft, u64 val) |
3327 | { | |
3328 | return -ENOSYS; | |
3329 | } | |
3330 | #endif | |
7dc74be0 | 3331 | |
406eb0c9 | 3332 | #ifdef CONFIG_NUMA |
2da8ca82 | 3333 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
406eb0c9 | 3334 | { |
25485de6 GT |
3335 | struct numa_stat { |
3336 | const char *name; | |
3337 | unsigned int lru_mask; | |
3338 | }; | |
3339 | ||
3340 | static const struct numa_stat stats[] = { | |
3341 | { "total", LRU_ALL }, | |
3342 | { "file", LRU_ALL_FILE }, | |
3343 | { "anon", LRU_ALL_ANON }, | |
3344 | { "unevictable", BIT(LRU_UNEVICTABLE) }, | |
3345 | }; | |
3346 | const struct numa_stat *stat; | |
406eb0c9 | 3347 | int nid; |
25485de6 | 3348 | unsigned long nr; |
aa9694bb | 3349 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
406eb0c9 | 3350 | |
25485de6 GT |
3351 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3352 | nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask); | |
3353 | seq_printf(m, "%s=%lu", stat->name, nr); | |
3354 | for_each_node_state(nid, N_MEMORY) { | |
3355 | nr = mem_cgroup_node_nr_lru_pages(memcg, nid, | |
3356 | stat->lru_mask); | |
3357 | seq_printf(m, " N%d=%lu", nid, nr); | |
3358 | } | |
3359 | seq_putc(m, '\n'); | |
406eb0c9 | 3360 | } |
406eb0c9 | 3361 | |
071aee13 YH |
3362 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3363 | struct mem_cgroup *iter; | |
3364 | ||
3365 | nr = 0; | |
3366 | for_each_mem_cgroup_tree(iter, memcg) | |
3367 | nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask); | |
3368 | seq_printf(m, "hierarchical_%s=%lu", stat->name, nr); | |
3369 | for_each_node_state(nid, N_MEMORY) { | |
3370 | nr = 0; | |
3371 | for_each_mem_cgroup_tree(iter, memcg) | |
3372 | nr += mem_cgroup_node_nr_lru_pages( | |
3373 | iter, nid, stat->lru_mask); | |
3374 | seq_printf(m, " N%d=%lu", nid, nr); | |
3375 | } | |
3376 | seq_putc(m, '\n'); | |
406eb0c9 | 3377 | } |
406eb0c9 | 3378 | |
406eb0c9 YH |
3379 | return 0; |
3380 | } | |
3381 | #endif /* CONFIG_NUMA */ | |
3382 | ||
df0e53d0 | 3383 | /* Universal VM events cgroup1 shows, original sort order */ |
8dd53fd3 | 3384 | static const unsigned int memcg1_events[] = { |
df0e53d0 JW |
3385 | PGPGIN, |
3386 | PGPGOUT, | |
3387 | PGFAULT, | |
3388 | PGMAJFAULT, | |
3389 | }; | |
3390 | ||
3391 | static const char *const memcg1_event_names[] = { | |
3392 | "pgpgin", | |
3393 | "pgpgout", | |
3394 | "pgfault", | |
3395 | "pgmajfault", | |
3396 | }; | |
3397 | ||
2da8ca82 | 3398 | static int memcg_stat_show(struct seq_file *m, void *v) |
d2ceb9b7 | 3399 | { |
aa9694bb | 3400 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
3e32cb2e | 3401 | unsigned long memory, memsw; |
af7c4b0e JW |
3402 | struct mem_cgroup *mi; |
3403 | unsigned int i; | |
8de7ecc6 | 3404 | struct accumulated_stats acc; |
406eb0c9 | 3405 | |
71cd3113 | 3406 | BUILD_BUG_ON(ARRAY_SIZE(memcg1_stat_names) != ARRAY_SIZE(memcg1_stats)); |
70bc068c RS |
3407 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); |
3408 | ||
71cd3113 JW |
3409 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
3410 | if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account()) | |
1dd3a273 | 3411 | continue; |
71cd3113 | 3412 | seq_printf(m, "%s %lu\n", memcg1_stat_names[i], |
ccda7f43 | 3413 | memcg_page_state(memcg, memcg1_stats[i]) * |
71cd3113 | 3414 | PAGE_SIZE); |
1dd3a273 | 3415 | } |
7b854121 | 3416 | |
df0e53d0 JW |
3417 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
3418 | seq_printf(m, "%s %lu\n", memcg1_event_names[i], | |
ccda7f43 | 3419 | memcg_sum_events(memcg, memcg1_events[i])); |
af7c4b0e JW |
3420 | |
3421 | for (i = 0; i < NR_LRU_LISTS; i++) | |
3422 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], | |
3423 | mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE); | |
3424 | ||
14067bb3 | 3425 | /* Hierarchical information */ |
3e32cb2e JW |
3426 | memory = memsw = PAGE_COUNTER_MAX; |
3427 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { | |
bbec2e15 RG |
3428 | memory = min(memory, mi->memory.max); |
3429 | memsw = min(memsw, mi->memsw.max); | |
fee7b548 | 3430 | } |
3e32cb2e JW |
3431 | seq_printf(m, "hierarchical_memory_limit %llu\n", |
3432 | (u64)memory * PAGE_SIZE); | |
7941d214 | 3433 | if (do_memsw_account()) |
3e32cb2e JW |
3434 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
3435 | (u64)memsw * PAGE_SIZE); | |
7f016ee8 | 3436 | |
8de7ecc6 SB |
3437 | memset(&acc, 0, sizeof(acc)); |
3438 | acc.stats_size = ARRAY_SIZE(memcg1_stats); | |
3439 | acc.stats_array = memcg1_stats; | |
3440 | acc.events_size = ARRAY_SIZE(memcg1_events); | |
3441 | acc.events_array = memcg1_events; | |
3442 | accumulate_memcg_tree(memcg, &acc); | |
af7c4b0e | 3443 | |
8de7ecc6 | 3444 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
71cd3113 | 3445 | if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account()) |
1dd3a273 | 3446 | continue; |
8de7ecc6 SB |
3447 | seq_printf(m, "total_%s %llu\n", memcg1_stat_names[i], |
3448 | (u64)acc.stat[i] * PAGE_SIZE); | |
af7c4b0e JW |
3449 | } |
3450 | ||
8de7ecc6 SB |
3451 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
3452 | seq_printf(m, "total_%s %llu\n", memcg1_event_names[i], | |
3453 | (u64)acc.events[i]); | |
af7c4b0e | 3454 | |
8de7ecc6 SB |
3455 | for (i = 0; i < NR_LRU_LISTS; i++) |
3456 | seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], | |
3457 | (u64)acc.lru_pages[i] * PAGE_SIZE); | |
14067bb3 | 3458 | |
7f016ee8 | 3459 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 | 3460 | { |
ef8f2327 MG |
3461 | pg_data_t *pgdat; |
3462 | struct mem_cgroup_per_node *mz; | |
89abfab1 | 3463 | struct zone_reclaim_stat *rstat; |
7f016ee8 KM |
3464 | unsigned long recent_rotated[2] = {0, 0}; |
3465 | unsigned long recent_scanned[2] = {0, 0}; | |
3466 | ||
ef8f2327 MG |
3467 | for_each_online_pgdat(pgdat) { |
3468 | mz = mem_cgroup_nodeinfo(memcg, pgdat->node_id); | |
3469 | rstat = &mz->lruvec.reclaim_stat; | |
7f016ee8 | 3470 | |
ef8f2327 MG |
3471 | recent_rotated[0] += rstat->recent_rotated[0]; |
3472 | recent_rotated[1] += rstat->recent_rotated[1]; | |
3473 | recent_scanned[0] += rstat->recent_scanned[0]; | |
3474 | recent_scanned[1] += rstat->recent_scanned[1]; | |
3475 | } | |
78ccf5b5 JW |
3476 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
3477 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); | |
3478 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); | |
3479 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); | |
7f016ee8 KM |
3480 | } |
3481 | #endif | |
3482 | ||
d2ceb9b7 KH |
3483 | return 0; |
3484 | } | |
3485 | ||
182446d0 TH |
3486 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
3487 | struct cftype *cft) | |
a7885eb8 | 3488 | { |
182446d0 | 3489 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3490 | |
1f4c025b | 3491 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
3492 | } |
3493 | ||
182446d0 TH |
3494 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
3495 | struct cftype *cft, u64 val) | |
a7885eb8 | 3496 | { |
182446d0 | 3497 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3498 | |
3dae7fec | 3499 | if (val > 100) |
a7885eb8 KM |
3500 | return -EINVAL; |
3501 | ||
14208b0e | 3502 | if (css->parent) |
3dae7fec JW |
3503 | memcg->swappiness = val; |
3504 | else | |
3505 | vm_swappiness = val; | |
068b38c1 | 3506 | |
a7885eb8 KM |
3507 | return 0; |
3508 | } | |
3509 | ||
2e72b634 KS |
3510 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
3511 | { | |
3512 | struct mem_cgroup_threshold_ary *t; | |
3e32cb2e | 3513 | unsigned long usage; |
2e72b634 KS |
3514 | int i; |
3515 | ||
3516 | rcu_read_lock(); | |
3517 | if (!swap) | |
2c488db2 | 3518 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 3519 | else |
2c488db2 | 3520 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
3521 | |
3522 | if (!t) | |
3523 | goto unlock; | |
3524 | ||
ce00a967 | 3525 | usage = mem_cgroup_usage(memcg, swap); |
2e72b634 KS |
3526 | |
3527 | /* | |
748dad36 | 3528 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
3529 | * If it's not true, a threshold was crossed after last |
3530 | * call of __mem_cgroup_threshold(). | |
3531 | */ | |
5407a562 | 3532 | i = t->current_threshold; |
2e72b634 KS |
3533 | |
3534 | /* | |
3535 | * Iterate backward over array of thresholds starting from | |
3536 | * current_threshold and check if a threshold is crossed. | |
3537 | * If none of thresholds below usage is crossed, we read | |
3538 | * only one element of the array here. | |
3539 | */ | |
3540 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
3541 | eventfd_signal(t->entries[i].eventfd, 1); | |
3542 | ||
3543 | /* i = current_threshold + 1 */ | |
3544 | i++; | |
3545 | ||
3546 | /* | |
3547 | * Iterate forward over array of thresholds starting from | |
3548 | * current_threshold+1 and check if a threshold is crossed. | |
3549 | * If none of thresholds above usage is crossed, we read | |
3550 | * only one element of the array here. | |
3551 | */ | |
3552 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
3553 | eventfd_signal(t->entries[i].eventfd, 1); | |
3554 | ||
3555 | /* Update current_threshold */ | |
5407a562 | 3556 | t->current_threshold = i - 1; |
2e72b634 KS |
3557 | unlock: |
3558 | rcu_read_unlock(); | |
3559 | } | |
3560 | ||
3561 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
3562 | { | |
ad4ca5f4 KS |
3563 | while (memcg) { |
3564 | __mem_cgroup_threshold(memcg, false); | |
7941d214 | 3565 | if (do_memsw_account()) |
ad4ca5f4 KS |
3566 | __mem_cgroup_threshold(memcg, true); |
3567 | ||
3568 | memcg = parent_mem_cgroup(memcg); | |
3569 | } | |
2e72b634 KS |
3570 | } |
3571 | ||
3572 | static int compare_thresholds(const void *a, const void *b) | |
3573 | { | |
3574 | const struct mem_cgroup_threshold *_a = a; | |
3575 | const struct mem_cgroup_threshold *_b = b; | |
3576 | ||
2bff24a3 GT |
3577 | if (_a->threshold > _b->threshold) |
3578 | return 1; | |
3579 | ||
3580 | if (_a->threshold < _b->threshold) | |
3581 | return -1; | |
3582 | ||
3583 | return 0; | |
2e72b634 KS |
3584 | } |
3585 | ||
c0ff4b85 | 3586 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
3587 | { |
3588 | struct mem_cgroup_eventfd_list *ev; | |
3589 | ||
2bcf2e92 MH |
3590 | spin_lock(&memcg_oom_lock); |
3591 | ||
c0ff4b85 | 3592 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 | 3593 | eventfd_signal(ev->eventfd, 1); |
2bcf2e92 MH |
3594 | |
3595 | spin_unlock(&memcg_oom_lock); | |
9490ff27 KH |
3596 | return 0; |
3597 | } | |
3598 | ||
c0ff4b85 | 3599 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 3600 | { |
7d74b06f KH |
3601 | struct mem_cgroup *iter; |
3602 | ||
c0ff4b85 | 3603 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 3604 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
3605 | } |
3606 | ||
59b6f873 | 3607 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 | 3608 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
2e72b634 | 3609 | { |
2c488db2 KS |
3610 | struct mem_cgroup_thresholds *thresholds; |
3611 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e JW |
3612 | unsigned long threshold; |
3613 | unsigned long usage; | |
2c488db2 | 3614 | int i, size, ret; |
2e72b634 | 3615 | |
650c5e56 | 3616 | ret = page_counter_memparse(args, "-1", &threshold); |
2e72b634 KS |
3617 | if (ret) |
3618 | return ret; | |
3619 | ||
3620 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 3621 | |
05b84301 | 3622 | if (type == _MEM) { |
2c488db2 | 3623 | thresholds = &memcg->thresholds; |
ce00a967 | 3624 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 3625 | } else if (type == _MEMSWAP) { |
2c488db2 | 3626 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 3627 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 3628 | } else |
2e72b634 KS |
3629 | BUG(); |
3630 | ||
2e72b634 | 3631 | /* Check if a threshold crossed before adding a new one */ |
2c488db2 | 3632 | if (thresholds->primary) |
2e72b634 KS |
3633 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
3634 | ||
2c488db2 | 3635 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
3636 | |
3637 | /* Allocate memory for new array of thresholds */ | |
67b8046f | 3638 | new = kmalloc(struct_size(new, entries, size), GFP_KERNEL); |
2c488db2 | 3639 | if (!new) { |
2e72b634 KS |
3640 | ret = -ENOMEM; |
3641 | goto unlock; | |
3642 | } | |
2c488db2 | 3643 | new->size = size; |
2e72b634 KS |
3644 | |
3645 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
3646 | if (thresholds->primary) { |
3647 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 3648 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
3649 | } |
3650 | ||
2e72b634 | 3651 | /* Add new threshold */ |
2c488db2 KS |
3652 | new->entries[size - 1].eventfd = eventfd; |
3653 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
3654 | |
3655 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 3656 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
3657 | compare_thresholds, NULL); |
3658 | ||
3659 | /* Find current threshold */ | |
2c488db2 | 3660 | new->current_threshold = -1; |
2e72b634 | 3661 | for (i = 0; i < size; i++) { |
748dad36 | 3662 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 3663 | /* |
2c488db2 KS |
3664 | * new->current_threshold will not be used until |
3665 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
3666 | * it here. |
3667 | */ | |
2c488db2 | 3668 | ++new->current_threshold; |
748dad36 SZ |
3669 | } else |
3670 | break; | |
2e72b634 KS |
3671 | } |
3672 | ||
2c488db2 KS |
3673 | /* Free old spare buffer and save old primary buffer as spare */ |
3674 | kfree(thresholds->spare); | |
3675 | thresholds->spare = thresholds->primary; | |
3676 | ||
3677 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 3678 | |
907860ed | 3679 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
3680 | synchronize_rcu(); |
3681 | ||
2e72b634 KS |
3682 | unlock: |
3683 | mutex_unlock(&memcg->thresholds_lock); | |
3684 | ||
3685 | return ret; | |
3686 | } | |
3687 | ||
59b6f873 | 3688 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3689 | struct eventfd_ctx *eventfd, const char *args) |
3690 | { | |
59b6f873 | 3691 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
347c4a87 TH |
3692 | } |
3693 | ||
59b6f873 | 3694 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3695 | struct eventfd_ctx *eventfd, const char *args) |
3696 | { | |
59b6f873 | 3697 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
347c4a87 TH |
3698 | } |
3699 | ||
59b6f873 | 3700 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 3701 | struct eventfd_ctx *eventfd, enum res_type type) |
2e72b634 | 3702 | { |
2c488db2 KS |
3703 | struct mem_cgroup_thresholds *thresholds; |
3704 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e | 3705 | unsigned long usage; |
2c488db2 | 3706 | int i, j, size; |
2e72b634 KS |
3707 | |
3708 | mutex_lock(&memcg->thresholds_lock); | |
05b84301 JW |
3709 | |
3710 | if (type == _MEM) { | |
2c488db2 | 3711 | thresholds = &memcg->thresholds; |
ce00a967 | 3712 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 3713 | } else if (type == _MEMSWAP) { |
2c488db2 | 3714 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 3715 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 3716 | } else |
2e72b634 KS |
3717 | BUG(); |
3718 | ||
371528ca AV |
3719 | if (!thresholds->primary) |
3720 | goto unlock; | |
3721 | ||
2e72b634 KS |
3722 | /* Check if a threshold crossed before removing */ |
3723 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
3724 | ||
3725 | /* Calculate new number of threshold */ | |
2c488db2 KS |
3726 | size = 0; |
3727 | for (i = 0; i < thresholds->primary->size; i++) { | |
3728 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
3729 | size++; |
3730 | } | |
3731 | ||
2c488db2 | 3732 | new = thresholds->spare; |
907860ed | 3733 | |
2e72b634 KS |
3734 | /* Set thresholds array to NULL if we don't have thresholds */ |
3735 | if (!size) { | |
2c488db2 KS |
3736 | kfree(new); |
3737 | new = NULL; | |
907860ed | 3738 | goto swap_buffers; |
2e72b634 KS |
3739 | } |
3740 | ||
2c488db2 | 3741 | new->size = size; |
2e72b634 KS |
3742 | |
3743 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
3744 | new->current_threshold = -1; |
3745 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
3746 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
3747 | continue; |
3748 | ||
2c488db2 | 3749 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 3750 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 3751 | /* |
2c488db2 | 3752 | * new->current_threshold will not be used |
2e72b634 KS |
3753 | * until rcu_assign_pointer(), so it's safe to increment |
3754 | * it here. | |
3755 | */ | |
2c488db2 | 3756 | ++new->current_threshold; |
2e72b634 KS |
3757 | } |
3758 | j++; | |
3759 | } | |
3760 | ||
907860ed | 3761 | swap_buffers: |
2c488db2 KS |
3762 | /* Swap primary and spare array */ |
3763 | thresholds->spare = thresholds->primary; | |
8c757763 | 3764 | |
2c488db2 | 3765 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 3766 | |
907860ed | 3767 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 3768 | synchronize_rcu(); |
6611d8d7 MC |
3769 | |
3770 | /* If all events are unregistered, free the spare array */ | |
3771 | if (!new) { | |
3772 | kfree(thresholds->spare); | |
3773 | thresholds->spare = NULL; | |
3774 | } | |
371528ca | 3775 | unlock: |
2e72b634 | 3776 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 3777 | } |
c1e862c1 | 3778 | |
59b6f873 | 3779 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3780 | struct eventfd_ctx *eventfd) |
3781 | { | |
59b6f873 | 3782 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
347c4a87 TH |
3783 | } |
3784 | ||
59b6f873 | 3785 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3786 | struct eventfd_ctx *eventfd) |
3787 | { | |
59b6f873 | 3788 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
347c4a87 TH |
3789 | } |
3790 | ||
59b6f873 | 3791 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
347c4a87 | 3792 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 3793 | { |
9490ff27 | 3794 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 3795 | |
9490ff27 KH |
3796 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
3797 | if (!event) | |
3798 | return -ENOMEM; | |
3799 | ||
1af8efe9 | 3800 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
3801 | |
3802 | event->eventfd = eventfd; | |
3803 | list_add(&event->list, &memcg->oom_notify); | |
3804 | ||
3805 | /* already in OOM ? */ | |
c2b42d3c | 3806 | if (memcg->under_oom) |
9490ff27 | 3807 | eventfd_signal(eventfd, 1); |
1af8efe9 | 3808 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
3809 | |
3810 | return 0; | |
3811 | } | |
3812 | ||
59b6f873 | 3813 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 3814 | struct eventfd_ctx *eventfd) |
9490ff27 | 3815 | { |
9490ff27 | 3816 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 3817 | |
1af8efe9 | 3818 | spin_lock(&memcg_oom_lock); |
9490ff27 | 3819 | |
c0ff4b85 | 3820 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
3821 | if (ev->eventfd == eventfd) { |
3822 | list_del(&ev->list); | |
3823 | kfree(ev); | |
3824 | } | |
3825 | } | |
3826 | ||
1af8efe9 | 3827 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
3828 | } |
3829 | ||
2da8ca82 | 3830 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
3c11ecf4 | 3831 | { |
aa9694bb | 3832 | struct mem_cgroup *memcg = mem_cgroup_from_seq(sf); |
3c11ecf4 | 3833 | |
791badbd | 3834 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
c2b42d3c | 3835 | seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom); |
fe6bdfc8 RG |
3836 | seq_printf(sf, "oom_kill %lu\n", |
3837 | atomic_long_read(&memcg->memory_events[MEMCG_OOM_KILL])); | |
3c11ecf4 KH |
3838 | return 0; |
3839 | } | |
3840 | ||
182446d0 | 3841 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
3842 | struct cftype *cft, u64 val) |
3843 | { | |
182446d0 | 3844 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 KH |
3845 | |
3846 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
14208b0e | 3847 | if (!css->parent || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
3848 | return -EINVAL; |
3849 | ||
c0ff4b85 | 3850 | memcg->oom_kill_disable = val; |
4d845ebf | 3851 | if (!val) |
c0ff4b85 | 3852 | memcg_oom_recover(memcg); |
3dae7fec | 3853 | |
3c11ecf4 KH |
3854 | return 0; |
3855 | } | |
3856 | ||
52ebea74 TH |
3857 | #ifdef CONFIG_CGROUP_WRITEBACK |
3858 | ||
841710aa TH |
3859 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) |
3860 | { | |
3861 | return wb_domain_init(&memcg->cgwb_domain, gfp); | |
3862 | } | |
3863 | ||
3864 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
3865 | { | |
3866 | wb_domain_exit(&memcg->cgwb_domain); | |
3867 | } | |
3868 | ||
2529bb3a TH |
3869 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
3870 | { | |
3871 | wb_domain_size_changed(&memcg->cgwb_domain); | |
3872 | } | |
3873 | ||
841710aa TH |
3874 | struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb) |
3875 | { | |
3876 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
3877 | ||
3878 | if (!memcg->css.parent) | |
3879 | return NULL; | |
3880 | ||
3881 | return &memcg->cgwb_domain; | |
3882 | } | |
3883 | ||
c2aa723a TH |
3884 | /** |
3885 | * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg | |
3886 | * @wb: bdi_writeback in question | |
c5edf9cd TH |
3887 | * @pfilepages: out parameter for number of file pages |
3888 | * @pheadroom: out parameter for number of allocatable pages according to memcg | |
c2aa723a TH |
3889 | * @pdirty: out parameter for number of dirty pages |
3890 | * @pwriteback: out parameter for number of pages under writeback | |
3891 | * | |
c5edf9cd TH |
3892 | * Determine the numbers of file, headroom, dirty, and writeback pages in |
3893 | * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom | |
3894 | * is a bit more involved. | |
c2aa723a | 3895 | * |
c5edf9cd TH |
3896 | * A memcg's headroom is "min(max, high) - used". In the hierarchy, the |
3897 | * headroom is calculated as the lowest headroom of itself and the | |
3898 | * ancestors. Note that this doesn't consider the actual amount of | |
3899 | * available memory in the system. The caller should further cap | |
3900 | * *@pheadroom accordingly. | |
c2aa723a | 3901 | */ |
c5edf9cd TH |
3902 | void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages, |
3903 | unsigned long *pheadroom, unsigned long *pdirty, | |
3904 | unsigned long *pwriteback) | |
c2aa723a TH |
3905 | { |
3906 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
3907 | struct mem_cgroup *parent; | |
c2aa723a | 3908 | |
ccda7f43 | 3909 | *pdirty = memcg_page_state(memcg, NR_FILE_DIRTY); |
c2aa723a TH |
3910 | |
3911 | /* this should eventually include NR_UNSTABLE_NFS */ | |
ccda7f43 | 3912 | *pwriteback = memcg_page_state(memcg, NR_WRITEBACK); |
c5edf9cd TH |
3913 | *pfilepages = mem_cgroup_nr_lru_pages(memcg, (1 << LRU_INACTIVE_FILE) | |
3914 | (1 << LRU_ACTIVE_FILE)); | |
3915 | *pheadroom = PAGE_COUNTER_MAX; | |
c2aa723a | 3916 | |
c2aa723a | 3917 | while ((parent = parent_mem_cgroup(memcg))) { |
bbec2e15 | 3918 | unsigned long ceiling = min(memcg->memory.max, memcg->high); |
c2aa723a TH |
3919 | unsigned long used = page_counter_read(&memcg->memory); |
3920 | ||
c5edf9cd | 3921 | *pheadroom = min(*pheadroom, ceiling - min(ceiling, used)); |
c2aa723a TH |
3922 | memcg = parent; |
3923 | } | |
c2aa723a TH |
3924 | } |
3925 | ||
841710aa TH |
3926 | #else /* CONFIG_CGROUP_WRITEBACK */ |
3927 | ||
3928 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) | |
3929 | { | |
3930 | return 0; | |
3931 | } | |
3932 | ||
3933 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
3934 | { | |
3935 | } | |
3936 | ||
2529bb3a TH |
3937 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
3938 | { | |
3939 | } | |
3940 | ||
52ebea74 TH |
3941 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
3942 | ||
3bc942f3 TH |
3943 | /* |
3944 | * DO NOT USE IN NEW FILES. | |
3945 | * | |
3946 | * "cgroup.event_control" implementation. | |
3947 | * | |
3948 | * This is way over-engineered. It tries to support fully configurable | |
3949 | * events for each user. Such level of flexibility is completely | |
3950 | * unnecessary especially in the light of the planned unified hierarchy. | |
3951 | * | |
3952 | * Please deprecate this and replace with something simpler if at all | |
3953 | * possible. | |
3954 | */ | |
3955 | ||
79bd9814 TH |
3956 | /* |
3957 | * Unregister event and free resources. | |
3958 | * | |
3959 | * Gets called from workqueue. | |
3960 | */ | |
3bc942f3 | 3961 | static void memcg_event_remove(struct work_struct *work) |
79bd9814 | 3962 | { |
3bc942f3 TH |
3963 | struct mem_cgroup_event *event = |
3964 | container_of(work, struct mem_cgroup_event, remove); | |
59b6f873 | 3965 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
3966 | |
3967 | remove_wait_queue(event->wqh, &event->wait); | |
3968 | ||
59b6f873 | 3969 | event->unregister_event(memcg, event->eventfd); |
79bd9814 TH |
3970 | |
3971 | /* Notify userspace the event is going away. */ | |
3972 | eventfd_signal(event->eventfd, 1); | |
3973 | ||
3974 | eventfd_ctx_put(event->eventfd); | |
3975 | kfree(event); | |
59b6f873 | 3976 | css_put(&memcg->css); |
79bd9814 TH |
3977 | } |
3978 | ||
3979 | /* | |
a9a08845 | 3980 | * Gets called on EPOLLHUP on eventfd when user closes it. |
79bd9814 TH |
3981 | * |
3982 | * Called with wqh->lock held and interrupts disabled. | |
3983 | */ | |
ac6424b9 | 3984 | static int memcg_event_wake(wait_queue_entry_t *wait, unsigned mode, |
3bc942f3 | 3985 | int sync, void *key) |
79bd9814 | 3986 | { |
3bc942f3 TH |
3987 | struct mem_cgroup_event *event = |
3988 | container_of(wait, struct mem_cgroup_event, wait); | |
59b6f873 | 3989 | struct mem_cgroup *memcg = event->memcg; |
3ad6f93e | 3990 | __poll_t flags = key_to_poll(key); |
79bd9814 | 3991 | |
a9a08845 | 3992 | if (flags & EPOLLHUP) { |
79bd9814 TH |
3993 | /* |
3994 | * If the event has been detached at cgroup removal, we | |
3995 | * can simply return knowing the other side will cleanup | |
3996 | * for us. | |
3997 | * | |
3998 | * We can't race against event freeing since the other | |
3999 | * side will require wqh->lock via remove_wait_queue(), | |
4000 | * which we hold. | |
4001 | */ | |
fba94807 | 4002 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
4003 | if (!list_empty(&event->list)) { |
4004 | list_del_init(&event->list); | |
4005 | /* | |
4006 | * We are in atomic context, but cgroup_event_remove() | |
4007 | * may sleep, so we have to call it in workqueue. | |
4008 | */ | |
4009 | schedule_work(&event->remove); | |
4010 | } | |
fba94807 | 4011 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
4012 | } |
4013 | ||
4014 | return 0; | |
4015 | } | |
4016 | ||
3bc942f3 | 4017 | static void memcg_event_ptable_queue_proc(struct file *file, |
79bd9814 TH |
4018 | wait_queue_head_t *wqh, poll_table *pt) |
4019 | { | |
3bc942f3 TH |
4020 | struct mem_cgroup_event *event = |
4021 | container_of(pt, struct mem_cgroup_event, pt); | |
79bd9814 TH |
4022 | |
4023 | event->wqh = wqh; | |
4024 | add_wait_queue(wqh, &event->wait); | |
4025 | } | |
4026 | ||
4027 | /* | |
3bc942f3 TH |
4028 | * DO NOT USE IN NEW FILES. |
4029 | * | |
79bd9814 TH |
4030 | * Parse input and register new cgroup event handler. |
4031 | * | |
4032 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
4033 | * Interpretation of args is defined by control file implementation. | |
4034 | */ | |
451af504 TH |
4035 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
4036 | char *buf, size_t nbytes, loff_t off) | |
79bd9814 | 4037 | { |
451af504 | 4038 | struct cgroup_subsys_state *css = of_css(of); |
fba94807 | 4039 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4040 | struct mem_cgroup_event *event; |
79bd9814 TH |
4041 | struct cgroup_subsys_state *cfile_css; |
4042 | unsigned int efd, cfd; | |
4043 | struct fd efile; | |
4044 | struct fd cfile; | |
fba94807 | 4045 | const char *name; |
79bd9814 TH |
4046 | char *endp; |
4047 | int ret; | |
4048 | ||
451af504 TH |
4049 | buf = strstrip(buf); |
4050 | ||
4051 | efd = simple_strtoul(buf, &endp, 10); | |
79bd9814 TH |
4052 | if (*endp != ' ') |
4053 | return -EINVAL; | |
451af504 | 4054 | buf = endp + 1; |
79bd9814 | 4055 | |
451af504 | 4056 | cfd = simple_strtoul(buf, &endp, 10); |
79bd9814 TH |
4057 | if ((*endp != ' ') && (*endp != '\0')) |
4058 | return -EINVAL; | |
451af504 | 4059 | buf = endp + 1; |
79bd9814 TH |
4060 | |
4061 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
4062 | if (!event) | |
4063 | return -ENOMEM; | |
4064 | ||
59b6f873 | 4065 | event->memcg = memcg; |
79bd9814 | 4066 | INIT_LIST_HEAD(&event->list); |
3bc942f3 TH |
4067 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
4068 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); | |
4069 | INIT_WORK(&event->remove, memcg_event_remove); | |
79bd9814 TH |
4070 | |
4071 | efile = fdget(efd); | |
4072 | if (!efile.file) { | |
4073 | ret = -EBADF; | |
4074 | goto out_kfree; | |
4075 | } | |
4076 | ||
4077 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
4078 | if (IS_ERR(event->eventfd)) { | |
4079 | ret = PTR_ERR(event->eventfd); | |
4080 | goto out_put_efile; | |
4081 | } | |
4082 | ||
4083 | cfile = fdget(cfd); | |
4084 | if (!cfile.file) { | |
4085 | ret = -EBADF; | |
4086 | goto out_put_eventfd; | |
4087 | } | |
4088 | ||
4089 | /* the process need read permission on control file */ | |
4090 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
4091 | ret = inode_permission(file_inode(cfile.file), MAY_READ); | |
4092 | if (ret < 0) | |
4093 | goto out_put_cfile; | |
4094 | ||
fba94807 TH |
4095 | /* |
4096 | * Determine the event callbacks and set them in @event. This used | |
4097 | * to be done via struct cftype but cgroup core no longer knows | |
4098 | * about these events. The following is crude but the whole thing | |
4099 | * is for compatibility anyway. | |
3bc942f3 TH |
4100 | * |
4101 | * DO NOT ADD NEW FILES. | |
fba94807 | 4102 | */ |
b583043e | 4103 | name = cfile.file->f_path.dentry->d_name.name; |
fba94807 TH |
4104 | |
4105 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
4106 | event->register_event = mem_cgroup_usage_register_event; | |
4107 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
4108 | } else if (!strcmp(name, "memory.oom_control")) { | |
4109 | event->register_event = mem_cgroup_oom_register_event; | |
4110 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
4111 | } else if (!strcmp(name, "memory.pressure_level")) { | |
4112 | event->register_event = vmpressure_register_event; | |
4113 | event->unregister_event = vmpressure_unregister_event; | |
4114 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
4115 | event->register_event = memsw_cgroup_usage_register_event; |
4116 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
4117 | } else { |
4118 | ret = -EINVAL; | |
4119 | goto out_put_cfile; | |
4120 | } | |
4121 | ||
79bd9814 | 4122 | /* |
b5557c4c TH |
4123 | * Verify @cfile should belong to @css. Also, remaining events are |
4124 | * automatically removed on cgroup destruction but the removal is | |
4125 | * asynchronous, so take an extra ref on @css. | |
79bd9814 | 4126 | */ |
b583043e | 4127 | cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent, |
ec903c0c | 4128 | &memory_cgrp_subsys); |
79bd9814 | 4129 | ret = -EINVAL; |
5a17f543 | 4130 | if (IS_ERR(cfile_css)) |
79bd9814 | 4131 | goto out_put_cfile; |
5a17f543 TH |
4132 | if (cfile_css != css) { |
4133 | css_put(cfile_css); | |
79bd9814 | 4134 | goto out_put_cfile; |
5a17f543 | 4135 | } |
79bd9814 | 4136 | |
451af504 | 4137 | ret = event->register_event(memcg, event->eventfd, buf); |
79bd9814 TH |
4138 | if (ret) |
4139 | goto out_put_css; | |
4140 | ||
9965ed17 | 4141 | vfs_poll(efile.file, &event->pt); |
79bd9814 | 4142 | |
fba94807 TH |
4143 | spin_lock(&memcg->event_list_lock); |
4144 | list_add(&event->list, &memcg->event_list); | |
4145 | spin_unlock(&memcg->event_list_lock); | |
79bd9814 TH |
4146 | |
4147 | fdput(cfile); | |
4148 | fdput(efile); | |
4149 | ||
451af504 | 4150 | return nbytes; |
79bd9814 TH |
4151 | |
4152 | out_put_css: | |
b5557c4c | 4153 | css_put(css); |
79bd9814 TH |
4154 | out_put_cfile: |
4155 | fdput(cfile); | |
4156 | out_put_eventfd: | |
4157 | eventfd_ctx_put(event->eventfd); | |
4158 | out_put_efile: | |
4159 | fdput(efile); | |
4160 | out_kfree: | |
4161 | kfree(event); | |
4162 | ||
4163 | return ret; | |
4164 | } | |
4165 | ||
241994ed | 4166 | static struct cftype mem_cgroup_legacy_files[] = { |
8cdea7c0 | 4167 | { |
0eea1030 | 4168 | .name = "usage_in_bytes", |
8c7c6e34 | 4169 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
791badbd | 4170 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4171 | }, |
c84872e1 PE |
4172 | { |
4173 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 4174 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
6770c64e | 4175 | .write = mem_cgroup_reset, |
791badbd | 4176 | .read_u64 = mem_cgroup_read_u64, |
c84872e1 | 4177 | }, |
8cdea7c0 | 4178 | { |
0eea1030 | 4179 | .name = "limit_in_bytes", |
8c7c6e34 | 4180 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
451af504 | 4181 | .write = mem_cgroup_write, |
791badbd | 4182 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4183 | }, |
296c81d8 BS |
4184 | { |
4185 | .name = "soft_limit_in_bytes", | |
4186 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
451af504 | 4187 | .write = mem_cgroup_write, |
791badbd | 4188 | .read_u64 = mem_cgroup_read_u64, |
296c81d8 | 4189 | }, |
8cdea7c0 BS |
4190 | { |
4191 | .name = "failcnt", | |
8c7c6e34 | 4192 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
6770c64e | 4193 | .write = mem_cgroup_reset, |
791badbd | 4194 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4195 | }, |
d2ceb9b7 KH |
4196 | { |
4197 | .name = "stat", | |
2da8ca82 | 4198 | .seq_show = memcg_stat_show, |
d2ceb9b7 | 4199 | }, |
c1e862c1 KH |
4200 | { |
4201 | .name = "force_empty", | |
6770c64e | 4202 | .write = mem_cgroup_force_empty_write, |
c1e862c1 | 4203 | }, |
18f59ea7 BS |
4204 | { |
4205 | .name = "use_hierarchy", | |
4206 | .write_u64 = mem_cgroup_hierarchy_write, | |
4207 | .read_u64 = mem_cgroup_hierarchy_read, | |
4208 | }, | |
79bd9814 | 4209 | { |
3bc942f3 | 4210 | .name = "cgroup.event_control", /* XXX: for compat */ |
451af504 | 4211 | .write = memcg_write_event_control, |
7dbdb199 | 4212 | .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE, |
79bd9814 | 4213 | }, |
a7885eb8 KM |
4214 | { |
4215 | .name = "swappiness", | |
4216 | .read_u64 = mem_cgroup_swappiness_read, | |
4217 | .write_u64 = mem_cgroup_swappiness_write, | |
4218 | }, | |
7dc74be0 DN |
4219 | { |
4220 | .name = "move_charge_at_immigrate", | |
4221 | .read_u64 = mem_cgroup_move_charge_read, | |
4222 | .write_u64 = mem_cgroup_move_charge_write, | |
4223 | }, | |
9490ff27 KH |
4224 | { |
4225 | .name = "oom_control", | |
2da8ca82 | 4226 | .seq_show = mem_cgroup_oom_control_read, |
3c11ecf4 | 4227 | .write_u64 = mem_cgroup_oom_control_write, |
9490ff27 KH |
4228 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
4229 | }, | |
70ddf637 AV |
4230 | { |
4231 | .name = "pressure_level", | |
70ddf637 | 4232 | }, |
406eb0c9 YH |
4233 | #ifdef CONFIG_NUMA |
4234 | { | |
4235 | .name = "numa_stat", | |
2da8ca82 | 4236 | .seq_show = memcg_numa_stat_show, |
406eb0c9 YH |
4237 | }, |
4238 | #endif | |
510fc4e1 GC |
4239 | { |
4240 | .name = "kmem.limit_in_bytes", | |
4241 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
451af504 | 4242 | .write = mem_cgroup_write, |
791badbd | 4243 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4244 | }, |
4245 | { | |
4246 | .name = "kmem.usage_in_bytes", | |
4247 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
791badbd | 4248 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4249 | }, |
4250 | { | |
4251 | .name = "kmem.failcnt", | |
4252 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6770c64e | 4253 | .write = mem_cgroup_reset, |
791badbd | 4254 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4255 | }, |
4256 | { | |
4257 | .name = "kmem.max_usage_in_bytes", | |
4258 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6770c64e | 4259 | .write = mem_cgroup_reset, |
791badbd | 4260 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 | 4261 | }, |
5b365771 | 4262 | #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG) |
749c5415 GC |
4263 | { |
4264 | .name = "kmem.slabinfo", | |
bc2791f8 TH |
4265 | .seq_start = memcg_slab_start, |
4266 | .seq_next = memcg_slab_next, | |
4267 | .seq_stop = memcg_slab_stop, | |
b047501c | 4268 | .seq_show = memcg_slab_show, |
749c5415 GC |
4269 | }, |
4270 | #endif | |
d55f90bf VD |
4271 | { |
4272 | .name = "kmem.tcp.limit_in_bytes", | |
4273 | .private = MEMFILE_PRIVATE(_TCP, RES_LIMIT), | |
4274 | .write = mem_cgroup_write, | |
4275 | .read_u64 = mem_cgroup_read_u64, | |
4276 | }, | |
4277 | { | |
4278 | .name = "kmem.tcp.usage_in_bytes", | |
4279 | .private = MEMFILE_PRIVATE(_TCP, RES_USAGE), | |
4280 | .read_u64 = mem_cgroup_read_u64, | |
4281 | }, | |
4282 | { | |
4283 | .name = "kmem.tcp.failcnt", | |
4284 | .private = MEMFILE_PRIVATE(_TCP, RES_FAILCNT), | |
4285 | .write = mem_cgroup_reset, | |
4286 | .read_u64 = mem_cgroup_read_u64, | |
4287 | }, | |
4288 | { | |
4289 | .name = "kmem.tcp.max_usage_in_bytes", | |
4290 | .private = MEMFILE_PRIVATE(_TCP, RES_MAX_USAGE), | |
4291 | .write = mem_cgroup_reset, | |
4292 | .read_u64 = mem_cgroup_read_u64, | |
4293 | }, | |
6bc10349 | 4294 | { }, /* terminate */ |
af36f906 | 4295 | }; |
8c7c6e34 | 4296 | |
73f576c0 JW |
4297 | /* |
4298 | * Private memory cgroup IDR | |
4299 | * | |
4300 | * Swap-out records and page cache shadow entries need to store memcg | |
4301 | * references in constrained space, so we maintain an ID space that is | |
4302 | * limited to 16 bit (MEM_CGROUP_ID_MAX), limiting the total number of | |
4303 | * memory-controlled cgroups to 64k. | |
4304 | * | |
4305 | * However, there usually are many references to the oflline CSS after | |
4306 | * the cgroup has been destroyed, such as page cache or reclaimable | |
4307 | * slab objects, that don't need to hang on to the ID. We want to keep | |
4308 | * those dead CSS from occupying IDs, or we might quickly exhaust the | |
4309 | * relatively small ID space and prevent the creation of new cgroups | |
4310 | * even when there are much fewer than 64k cgroups - possibly none. | |
4311 | * | |
4312 | * Maintain a private 16-bit ID space for memcg, and allow the ID to | |
4313 | * be freed and recycled when it's no longer needed, which is usually | |
4314 | * when the CSS is offlined. | |
4315 | * | |
4316 | * The only exception to that are records of swapped out tmpfs/shmem | |
4317 | * pages that need to be attributed to live ancestors on swapin. But | |
4318 | * those references are manageable from userspace. | |
4319 | */ | |
4320 | ||
4321 | static DEFINE_IDR(mem_cgroup_idr); | |
4322 | ||
7e97de0b KT |
4323 | static void mem_cgroup_id_remove(struct mem_cgroup *memcg) |
4324 | { | |
4325 | if (memcg->id.id > 0) { | |
4326 | idr_remove(&mem_cgroup_idr, memcg->id.id); | |
4327 | memcg->id.id = 0; | |
4328 | } | |
4329 | } | |
4330 | ||
615d66c3 | 4331 | static void mem_cgroup_id_get_many(struct mem_cgroup *memcg, unsigned int n) |
73f576c0 | 4332 | { |
1c2d479a | 4333 | refcount_add(n, &memcg->id.ref); |
73f576c0 JW |
4334 | } |
4335 | ||
615d66c3 | 4336 | static void mem_cgroup_id_put_many(struct mem_cgroup *memcg, unsigned int n) |
73f576c0 | 4337 | { |
1c2d479a | 4338 | if (refcount_sub_and_test(n, &memcg->id.ref)) { |
7e97de0b | 4339 | mem_cgroup_id_remove(memcg); |
73f576c0 JW |
4340 | |
4341 | /* Memcg ID pins CSS */ | |
4342 | css_put(&memcg->css); | |
4343 | } | |
4344 | } | |
4345 | ||
615d66c3 VD |
4346 | static inline void mem_cgroup_id_get(struct mem_cgroup *memcg) |
4347 | { | |
4348 | mem_cgroup_id_get_many(memcg, 1); | |
4349 | } | |
4350 | ||
4351 | static inline void mem_cgroup_id_put(struct mem_cgroup *memcg) | |
4352 | { | |
4353 | mem_cgroup_id_put_many(memcg, 1); | |
4354 | } | |
4355 | ||
73f576c0 JW |
4356 | /** |
4357 | * mem_cgroup_from_id - look up a memcg from a memcg id | |
4358 | * @id: the memcg id to look up | |
4359 | * | |
4360 | * Caller must hold rcu_read_lock(). | |
4361 | */ | |
4362 | struct mem_cgroup *mem_cgroup_from_id(unsigned short id) | |
4363 | { | |
4364 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
4365 | return idr_find(&mem_cgroup_idr, id); | |
4366 | } | |
4367 | ||
ef8f2327 | 4368 | static int alloc_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
4369 | { |
4370 | struct mem_cgroup_per_node *pn; | |
ef8f2327 | 4371 | int tmp = node; |
1ecaab2b KH |
4372 | /* |
4373 | * This routine is called against possible nodes. | |
4374 | * But it's BUG to call kmalloc() against offline node. | |
4375 | * | |
4376 | * TODO: this routine can waste much memory for nodes which will | |
4377 | * never be onlined. It's better to use memory hotplug callback | |
4378 | * function. | |
4379 | */ | |
41e3355d KH |
4380 | if (!node_state(node, N_NORMAL_MEMORY)) |
4381 | tmp = -1; | |
17295c88 | 4382 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
4383 | if (!pn) |
4384 | return 1; | |
1ecaab2b | 4385 | |
a983b5eb JW |
4386 | pn->lruvec_stat_cpu = alloc_percpu(struct lruvec_stat); |
4387 | if (!pn->lruvec_stat_cpu) { | |
00f3ca2c JW |
4388 | kfree(pn); |
4389 | return 1; | |
4390 | } | |
4391 | ||
ef8f2327 MG |
4392 | lruvec_init(&pn->lruvec); |
4393 | pn->usage_in_excess = 0; | |
4394 | pn->on_tree = false; | |
4395 | pn->memcg = memcg; | |
4396 | ||
54f72fe0 | 4397 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
4398 | return 0; |
4399 | } | |
4400 | ||
ef8f2327 | 4401 | static void free_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 4402 | { |
00f3ca2c JW |
4403 | struct mem_cgroup_per_node *pn = memcg->nodeinfo[node]; |
4404 | ||
4eaf431f MH |
4405 | if (!pn) |
4406 | return; | |
4407 | ||
a983b5eb | 4408 | free_percpu(pn->lruvec_stat_cpu); |
00f3ca2c | 4409 | kfree(pn); |
1ecaab2b KH |
4410 | } |
4411 | ||
40e952f9 | 4412 | static void __mem_cgroup_free(struct mem_cgroup *memcg) |
59927fb9 | 4413 | { |
c8b2a36f | 4414 | int node; |
59927fb9 | 4415 | |
c8b2a36f | 4416 | for_each_node(node) |
ef8f2327 | 4417 | free_mem_cgroup_per_node_info(memcg, node); |
a983b5eb | 4418 | free_percpu(memcg->stat_cpu); |
8ff69e2c | 4419 | kfree(memcg); |
59927fb9 | 4420 | } |
3afe36b1 | 4421 | |
40e952f9 TE |
4422 | static void mem_cgroup_free(struct mem_cgroup *memcg) |
4423 | { | |
4424 | memcg_wb_domain_exit(memcg); | |
4425 | __mem_cgroup_free(memcg); | |
4426 | } | |
4427 | ||
0b8f73e1 | 4428 | static struct mem_cgroup *mem_cgroup_alloc(void) |
8cdea7c0 | 4429 | { |
d142e3e6 | 4430 | struct mem_cgroup *memcg; |
0b8f73e1 | 4431 | size_t size; |
6d12e2d8 | 4432 | int node; |
8cdea7c0 | 4433 | |
0b8f73e1 JW |
4434 | size = sizeof(struct mem_cgroup); |
4435 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); | |
4436 | ||
4437 | memcg = kzalloc(size, GFP_KERNEL); | |
c0ff4b85 | 4438 | if (!memcg) |
0b8f73e1 JW |
4439 | return NULL; |
4440 | ||
73f576c0 JW |
4441 | memcg->id.id = idr_alloc(&mem_cgroup_idr, NULL, |
4442 | 1, MEM_CGROUP_ID_MAX, | |
4443 | GFP_KERNEL); | |
4444 | if (memcg->id.id < 0) | |
4445 | goto fail; | |
4446 | ||
a983b5eb JW |
4447 | memcg->stat_cpu = alloc_percpu(struct mem_cgroup_stat_cpu); |
4448 | if (!memcg->stat_cpu) | |
0b8f73e1 | 4449 | goto fail; |
78fb7466 | 4450 | |
3ed28fa1 | 4451 | for_each_node(node) |
ef8f2327 | 4452 | if (alloc_mem_cgroup_per_node_info(memcg, node)) |
0b8f73e1 | 4453 | goto fail; |
f64c3f54 | 4454 | |
0b8f73e1 JW |
4455 | if (memcg_wb_domain_init(memcg, GFP_KERNEL)) |
4456 | goto fail; | |
28dbc4b6 | 4457 | |
f7e1cb6e | 4458 | INIT_WORK(&memcg->high_work, high_work_func); |
d142e3e6 GC |
4459 | memcg->last_scanned_node = MAX_NUMNODES; |
4460 | INIT_LIST_HEAD(&memcg->oom_notify); | |
d142e3e6 GC |
4461 | mutex_init(&memcg->thresholds_lock); |
4462 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 4463 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
4464 | INIT_LIST_HEAD(&memcg->event_list); |
4465 | spin_lock_init(&memcg->event_list_lock); | |
d886f4e4 | 4466 | memcg->socket_pressure = jiffies; |
84c07d11 | 4467 | #ifdef CONFIG_MEMCG_KMEM |
900a38f0 | 4468 | memcg->kmemcg_id = -1; |
900a38f0 | 4469 | #endif |
52ebea74 TH |
4470 | #ifdef CONFIG_CGROUP_WRITEBACK |
4471 | INIT_LIST_HEAD(&memcg->cgwb_list); | |
4472 | #endif | |
73f576c0 | 4473 | idr_replace(&mem_cgroup_idr, memcg, memcg->id.id); |
0b8f73e1 JW |
4474 | return memcg; |
4475 | fail: | |
7e97de0b | 4476 | mem_cgroup_id_remove(memcg); |
40e952f9 | 4477 | __mem_cgroup_free(memcg); |
0b8f73e1 | 4478 | return NULL; |
d142e3e6 GC |
4479 | } |
4480 | ||
0b8f73e1 JW |
4481 | static struct cgroup_subsys_state * __ref |
4482 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
d142e3e6 | 4483 | { |
0b8f73e1 JW |
4484 | struct mem_cgroup *parent = mem_cgroup_from_css(parent_css); |
4485 | struct mem_cgroup *memcg; | |
4486 | long error = -ENOMEM; | |
d142e3e6 | 4487 | |
0b8f73e1 JW |
4488 | memcg = mem_cgroup_alloc(); |
4489 | if (!memcg) | |
4490 | return ERR_PTR(error); | |
d142e3e6 | 4491 | |
0b8f73e1 JW |
4492 | memcg->high = PAGE_COUNTER_MAX; |
4493 | memcg->soft_limit = PAGE_COUNTER_MAX; | |
4494 | if (parent) { | |
4495 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
4496 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
4497 | } | |
4498 | if (parent && parent->use_hierarchy) { | |
4499 | memcg->use_hierarchy = true; | |
3e32cb2e | 4500 | page_counter_init(&memcg->memory, &parent->memory); |
37e84351 | 4501 | page_counter_init(&memcg->swap, &parent->swap); |
3e32cb2e JW |
4502 | page_counter_init(&memcg->memsw, &parent->memsw); |
4503 | page_counter_init(&memcg->kmem, &parent->kmem); | |
0db15298 | 4504 | page_counter_init(&memcg->tcpmem, &parent->tcpmem); |
18f59ea7 | 4505 | } else { |
3e32cb2e | 4506 | page_counter_init(&memcg->memory, NULL); |
37e84351 | 4507 | page_counter_init(&memcg->swap, NULL); |
3e32cb2e JW |
4508 | page_counter_init(&memcg->memsw, NULL); |
4509 | page_counter_init(&memcg->kmem, NULL); | |
0db15298 | 4510 | page_counter_init(&memcg->tcpmem, NULL); |
8c7f6edb TH |
4511 | /* |
4512 | * Deeper hierachy with use_hierarchy == false doesn't make | |
4513 | * much sense so let cgroup subsystem know about this | |
4514 | * unfortunate state in our controller. | |
4515 | */ | |
d142e3e6 | 4516 | if (parent != root_mem_cgroup) |
073219e9 | 4517 | memory_cgrp_subsys.broken_hierarchy = true; |
18f59ea7 | 4518 | } |
d6441637 | 4519 | |
0b8f73e1 JW |
4520 | /* The following stuff does not apply to the root */ |
4521 | if (!parent) { | |
4522 | root_mem_cgroup = memcg; | |
4523 | return &memcg->css; | |
4524 | } | |
4525 | ||
b313aeee | 4526 | error = memcg_online_kmem(memcg); |
0b8f73e1 JW |
4527 | if (error) |
4528 | goto fail; | |
127424c8 | 4529 | |
f7e1cb6e | 4530 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 4531 | static_branch_inc(&memcg_sockets_enabled_key); |
f7e1cb6e | 4532 | |
0b8f73e1 JW |
4533 | return &memcg->css; |
4534 | fail: | |
7e97de0b | 4535 | mem_cgroup_id_remove(memcg); |
0b8f73e1 | 4536 | mem_cgroup_free(memcg); |
ea3a9645 | 4537 | return ERR_PTR(-ENOMEM); |
0b8f73e1 JW |
4538 | } |
4539 | ||
73f576c0 | 4540 | static int mem_cgroup_css_online(struct cgroup_subsys_state *css) |
0b8f73e1 | 4541 | { |
58fa2a55 VD |
4542 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
4543 | ||
0a4465d3 KT |
4544 | /* |
4545 | * A memcg must be visible for memcg_expand_shrinker_maps() | |
4546 | * by the time the maps are allocated. So, we allocate maps | |
4547 | * here, when for_each_mem_cgroup() can't skip it. | |
4548 | */ | |
4549 | if (memcg_alloc_shrinker_maps(memcg)) { | |
4550 | mem_cgroup_id_remove(memcg); | |
4551 | return -ENOMEM; | |
4552 | } | |
4553 | ||
73f576c0 | 4554 | /* Online state pins memcg ID, memcg ID pins CSS */ |
1c2d479a | 4555 | refcount_set(&memcg->id.ref, 1); |
73f576c0 | 4556 | css_get(css); |
2f7dd7a4 | 4557 | return 0; |
8cdea7c0 BS |
4558 | } |
4559 | ||
eb95419b | 4560 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 4561 | { |
eb95419b | 4562 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4563 | struct mem_cgroup_event *event, *tmp; |
79bd9814 TH |
4564 | |
4565 | /* | |
4566 | * Unregister events and notify userspace. | |
4567 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
4568 | * directory to avoid race between userspace and kernelspace. | |
4569 | */ | |
fba94807 TH |
4570 | spin_lock(&memcg->event_list_lock); |
4571 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { | |
79bd9814 TH |
4572 | list_del_init(&event->list); |
4573 | schedule_work(&event->remove); | |
4574 | } | |
fba94807 | 4575 | spin_unlock(&memcg->event_list_lock); |
ec64f515 | 4576 | |
bf8d5d52 | 4577 | page_counter_set_min(&memcg->memory, 0); |
23067153 | 4578 | page_counter_set_low(&memcg->memory, 0); |
63677c74 | 4579 | |
567e9ab2 | 4580 | memcg_offline_kmem(memcg); |
52ebea74 | 4581 | wb_memcg_offline(memcg); |
73f576c0 | 4582 | |
591edfb1 RG |
4583 | drain_all_stock(memcg); |
4584 | ||
73f576c0 | 4585 | mem_cgroup_id_put(memcg); |
df878fb0 KH |
4586 | } |
4587 | ||
6df38689 VD |
4588 | static void mem_cgroup_css_released(struct cgroup_subsys_state *css) |
4589 | { | |
4590 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
4591 | ||
4592 | invalidate_reclaim_iterators(memcg); | |
4593 | } | |
4594 | ||
eb95419b | 4595 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 4596 | { |
eb95419b | 4597 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
c268e994 | 4598 | |
f7e1cb6e | 4599 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 4600 | static_branch_dec(&memcg_sockets_enabled_key); |
127424c8 | 4601 | |
0db15298 | 4602 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active) |
d55f90bf | 4603 | static_branch_dec(&memcg_sockets_enabled_key); |
3893e302 | 4604 | |
0b8f73e1 JW |
4605 | vmpressure_cleanup(&memcg->vmpressure); |
4606 | cancel_work_sync(&memcg->high_work); | |
4607 | mem_cgroup_remove_from_trees(memcg); | |
0a4465d3 | 4608 | memcg_free_shrinker_maps(memcg); |
d886f4e4 | 4609 | memcg_free_kmem(memcg); |
0b8f73e1 | 4610 | mem_cgroup_free(memcg); |
8cdea7c0 BS |
4611 | } |
4612 | ||
1ced953b TH |
4613 | /** |
4614 | * mem_cgroup_css_reset - reset the states of a mem_cgroup | |
4615 | * @css: the target css | |
4616 | * | |
4617 | * Reset the states of the mem_cgroup associated with @css. This is | |
4618 | * invoked when the userland requests disabling on the default hierarchy | |
4619 | * but the memcg is pinned through dependency. The memcg should stop | |
4620 | * applying policies and should revert to the vanilla state as it may be | |
4621 | * made visible again. | |
4622 | * | |
4623 | * The current implementation only resets the essential configurations. | |
4624 | * This needs to be expanded to cover all the visible parts. | |
4625 | */ | |
4626 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) | |
4627 | { | |
4628 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
4629 | ||
bbec2e15 RG |
4630 | page_counter_set_max(&memcg->memory, PAGE_COUNTER_MAX); |
4631 | page_counter_set_max(&memcg->swap, PAGE_COUNTER_MAX); | |
4632 | page_counter_set_max(&memcg->memsw, PAGE_COUNTER_MAX); | |
4633 | page_counter_set_max(&memcg->kmem, PAGE_COUNTER_MAX); | |
4634 | page_counter_set_max(&memcg->tcpmem, PAGE_COUNTER_MAX); | |
bf8d5d52 | 4635 | page_counter_set_min(&memcg->memory, 0); |
23067153 | 4636 | page_counter_set_low(&memcg->memory, 0); |
241994ed | 4637 | memcg->high = PAGE_COUNTER_MAX; |
24d404dc | 4638 | memcg->soft_limit = PAGE_COUNTER_MAX; |
2529bb3a | 4639 | memcg_wb_domain_size_changed(memcg); |
1ced953b TH |
4640 | } |
4641 | ||
02491447 | 4642 | #ifdef CONFIG_MMU |
7dc74be0 | 4643 | /* Handlers for move charge at task migration. */ |
854ffa8d | 4644 | static int mem_cgroup_do_precharge(unsigned long count) |
7dc74be0 | 4645 | { |
05b84301 | 4646 | int ret; |
9476db97 | 4647 | |
d0164adc MG |
4648 | /* Try a single bulk charge without reclaim first, kswapd may wake */ |
4649 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count); | |
9476db97 | 4650 | if (!ret) { |
854ffa8d | 4651 | mc.precharge += count; |
854ffa8d DN |
4652 | return ret; |
4653 | } | |
9476db97 | 4654 | |
3674534b | 4655 | /* Try charges one by one with reclaim, but do not retry */ |
854ffa8d | 4656 | while (count--) { |
3674534b | 4657 | ret = try_charge(mc.to, GFP_KERNEL | __GFP_NORETRY, 1); |
38c5d72f | 4658 | if (ret) |
38c5d72f | 4659 | return ret; |
854ffa8d | 4660 | mc.precharge++; |
9476db97 | 4661 | cond_resched(); |
854ffa8d | 4662 | } |
9476db97 | 4663 | return 0; |
4ffef5fe DN |
4664 | } |
4665 | ||
4ffef5fe DN |
4666 | union mc_target { |
4667 | struct page *page; | |
02491447 | 4668 | swp_entry_t ent; |
4ffef5fe DN |
4669 | }; |
4670 | ||
4ffef5fe | 4671 | enum mc_target_type { |
8d32ff84 | 4672 | MC_TARGET_NONE = 0, |
4ffef5fe | 4673 | MC_TARGET_PAGE, |
02491447 | 4674 | MC_TARGET_SWAP, |
c733a828 | 4675 | MC_TARGET_DEVICE, |
4ffef5fe DN |
4676 | }; |
4677 | ||
90254a65 DN |
4678 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
4679 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 4680 | { |
c733a828 | 4681 | struct page *page = _vm_normal_page(vma, addr, ptent, true); |
4ffef5fe | 4682 | |
90254a65 DN |
4683 | if (!page || !page_mapped(page)) |
4684 | return NULL; | |
4685 | if (PageAnon(page)) { | |
1dfab5ab | 4686 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 4687 | return NULL; |
1dfab5ab JW |
4688 | } else { |
4689 | if (!(mc.flags & MOVE_FILE)) | |
4690 | return NULL; | |
4691 | } | |
90254a65 DN |
4692 | if (!get_page_unless_zero(page)) |
4693 | return NULL; | |
4694 | ||
4695 | return page; | |
4696 | } | |
4697 | ||
c733a828 | 4698 | #if defined(CONFIG_SWAP) || defined(CONFIG_DEVICE_PRIVATE) |
90254a65 | 4699 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
48406ef8 | 4700 | pte_t ptent, swp_entry_t *entry) |
90254a65 | 4701 | { |
90254a65 DN |
4702 | struct page *page = NULL; |
4703 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
4704 | ||
1dfab5ab | 4705 | if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent)) |
90254a65 | 4706 | return NULL; |
c733a828 JG |
4707 | |
4708 | /* | |
4709 | * Handle MEMORY_DEVICE_PRIVATE which are ZONE_DEVICE page belonging to | |
4710 | * a device and because they are not accessible by CPU they are store | |
4711 | * as special swap entry in the CPU page table. | |
4712 | */ | |
4713 | if (is_device_private_entry(ent)) { | |
4714 | page = device_private_entry_to_page(ent); | |
4715 | /* | |
4716 | * MEMORY_DEVICE_PRIVATE means ZONE_DEVICE page and which have | |
4717 | * a refcount of 1 when free (unlike normal page) | |
4718 | */ | |
4719 | if (!page_ref_add_unless(page, 1, 1)) | |
4720 | return NULL; | |
4721 | return page; | |
4722 | } | |
4723 | ||
4b91355e KH |
4724 | /* |
4725 | * Because lookup_swap_cache() updates some statistics counter, | |
4726 | * we call find_get_page() with swapper_space directly. | |
4727 | */ | |
f6ab1f7f | 4728 | page = find_get_page(swap_address_space(ent), swp_offset(ent)); |
7941d214 | 4729 | if (do_memsw_account()) |
90254a65 DN |
4730 | entry->val = ent.val; |
4731 | ||
4732 | return page; | |
4733 | } | |
4b91355e KH |
4734 | #else |
4735 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
48406ef8 | 4736 | pte_t ptent, swp_entry_t *entry) |
4b91355e KH |
4737 | { |
4738 | return NULL; | |
4739 | } | |
4740 | #endif | |
90254a65 | 4741 | |
87946a72 DN |
4742 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
4743 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4744 | { | |
4745 | struct page *page = NULL; | |
87946a72 DN |
4746 | struct address_space *mapping; |
4747 | pgoff_t pgoff; | |
4748 | ||
4749 | if (!vma->vm_file) /* anonymous vma */ | |
4750 | return NULL; | |
1dfab5ab | 4751 | if (!(mc.flags & MOVE_FILE)) |
87946a72 DN |
4752 | return NULL; |
4753 | ||
87946a72 | 4754 | mapping = vma->vm_file->f_mapping; |
0661a336 | 4755 | pgoff = linear_page_index(vma, addr); |
87946a72 DN |
4756 | |
4757 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
4758 | #ifdef CONFIG_SWAP |
4759 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
139b6a6f JW |
4760 | if (shmem_mapping(mapping)) { |
4761 | page = find_get_entry(mapping, pgoff); | |
3159f943 | 4762 | if (xa_is_value(page)) { |
139b6a6f | 4763 | swp_entry_t swp = radix_to_swp_entry(page); |
7941d214 | 4764 | if (do_memsw_account()) |
139b6a6f | 4765 | *entry = swp; |
f6ab1f7f HY |
4766 | page = find_get_page(swap_address_space(swp), |
4767 | swp_offset(swp)); | |
139b6a6f JW |
4768 | } |
4769 | } else | |
4770 | page = find_get_page(mapping, pgoff); | |
4771 | #else | |
4772 | page = find_get_page(mapping, pgoff); | |
aa3b1895 | 4773 | #endif |
87946a72 DN |
4774 | return page; |
4775 | } | |
4776 | ||
b1b0deab CG |
4777 | /** |
4778 | * mem_cgroup_move_account - move account of the page | |
4779 | * @page: the page | |
25843c2b | 4780 | * @compound: charge the page as compound or small page |
b1b0deab CG |
4781 | * @from: mem_cgroup which the page is moved from. |
4782 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
4783 | * | |
3ac808fd | 4784 | * The caller must make sure the page is not on LRU (isolate_page() is useful.) |
b1b0deab CG |
4785 | * |
4786 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" | |
4787 | * from old cgroup. | |
4788 | */ | |
4789 | static int mem_cgroup_move_account(struct page *page, | |
f627c2f5 | 4790 | bool compound, |
b1b0deab CG |
4791 | struct mem_cgroup *from, |
4792 | struct mem_cgroup *to) | |
4793 | { | |
4794 | unsigned long flags; | |
f627c2f5 | 4795 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
b1b0deab | 4796 | int ret; |
c4843a75 | 4797 | bool anon; |
b1b0deab CG |
4798 | |
4799 | VM_BUG_ON(from == to); | |
4800 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
f627c2f5 | 4801 | VM_BUG_ON(compound && !PageTransHuge(page)); |
b1b0deab CG |
4802 | |
4803 | /* | |
6a93ca8f | 4804 | * Prevent mem_cgroup_migrate() from looking at |
45637bab | 4805 | * page->mem_cgroup of its source page while we change it. |
b1b0deab | 4806 | */ |
f627c2f5 | 4807 | ret = -EBUSY; |
b1b0deab CG |
4808 | if (!trylock_page(page)) |
4809 | goto out; | |
4810 | ||
4811 | ret = -EINVAL; | |
4812 | if (page->mem_cgroup != from) | |
4813 | goto out_unlock; | |
4814 | ||
c4843a75 GT |
4815 | anon = PageAnon(page); |
4816 | ||
b1b0deab CG |
4817 | spin_lock_irqsave(&from->move_lock, flags); |
4818 | ||
c4843a75 | 4819 | if (!anon && page_mapped(page)) { |
c9019e9b JW |
4820 | __mod_memcg_state(from, NR_FILE_MAPPED, -nr_pages); |
4821 | __mod_memcg_state(to, NR_FILE_MAPPED, nr_pages); | |
b1b0deab CG |
4822 | } |
4823 | ||
c4843a75 GT |
4824 | /* |
4825 | * move_lock grabbed above and caller set from->moving_account, so | |
ccda7f43 | 4826 | * mod_memcg_page_state will serialize updates to PageDirty. |
c4843a75 GT |
4827 | * So mapping should be stable for dirty pages. |
4828 | */ | |
4829 | if (!anon && PageDirty(page)) { | |
4830 | struct address_space *mapping = page_mapping(page); | |
4831 | ||
4832 | if (mapping_cap_account_dirty(mapping)) { | |
c9019e9b JW |
4833 | __mod_memcg_state(from, NR_FILE_DIRTY, -nr_pages); |
4834 | __mod_memcg_state(to, NR_FILE_DIRTY, nr_pages); | |
c4843a75 GT |
4835 | } |
4836 | } | |
4837 | ||
b1b0deab | 4838 | if (PageWriteback(page)) { |
c9019e9b JW |
4839 | __mod_memcg_state(from, NR_WRITEBACK, -nr_pages); |
4840 | __mod_memcg_state(to, NR_WRITEBACK, nr_pages); | |
b1b0deab CG |
4841 | } |
4842 | ||
4843 | /* | |
4844 | * It is safe to change page->mem_cgroup here because the page | |
4845 | * is referenced, charged, and isolated - we can't race with | |
4846 | * uncharging, charging, migration, or LRU putback. | |
4847 | */ | |
4848 | ||
4849 | /* caller should have done css_get */ | |
4850 | page->mem_cgroup = to; | |
4851 | spin_unlock_irqrestore(&from->move_lock, flags); | |
4852 | ||
4853 | ret = 0; | |
4854 | ||
4855 | local_irq_disable(); | |
f627c2f5 | 4856 | mem_cgroup_charge_statistics(to, page, compound, nr_pages); |
b1b0deab | 4857 | memcg_check_events(to, page); |
f627c2f5 | 4858 | mem_cgroup_charge_statistics(from, page, compound, -nr_pages); |
b1b0deab CG |
4859 | memcg_check_events(from, page); |
4860 | local_irq_enable(); | |
4861 | out_unlock: | |
4862 | unlock_page(page); | |
4863 | out: | |
4864 | return ret; | |
4865 | } | |
4866 | ||
7cf7806c LR |
4867 | /** |
4868 | * get_mctgt_type - get target type of moving charge | |
4869 | * @vma: the vma the pte to be checked belongs | |
4870 | * @addr: the address corresponding to the pte to be checked | |
4871 | * @ptent: the pte to be checked | |
4872 | * @target: the pointer the target page or swap ent will be stored(can be NULL) | |
4873 | * | |
4874 | * Returns | |
4875 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
4876 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
4877 | * move charge. if @target is not NULL, the page is stored in target->page | |
4878 | * with extra refcnt got(Callers should handle it). | |
4879 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a | |
4880 | * target for charge migration. if @target is not NULL, the entry is stored | |
4881 | * in target->ent. | |
df6ad698 JG |
4882 | * 3(MC_TARGET_DEVICE): like MC_TARGET_PAGE but page is MEMORY_DEVICE_PUBLIC |
4883 | * or MEMORY_DEVICE_PRIVATE (so ZONE_DEVICE page and thus not on the lru). | |
4884 | * For now we such page is charge like a regular page would be as for all | |
4885 | * intent and purposes it is just special memory taking the place of a | |
4886 | * regular page. | |
c733a828 JG |
4887 | * |
4888 | * See Documentations/vm/hmm.txt and include/linux/hmm.h | |
7cf7806c LR |
4889 | * |
4890 | * Called with pte lock held. | |
4891 | */ | |
4892 | ||
8d32ff84 | 4893 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
4894 | unsigned long addr, pte_t ptent, union mc_target *target) |
4895 | { | |
4896 | struct page *page = NULL; | |
8d32ff84 | 4897 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
4898 | swp_entry_t ent = { .val = 0 }; |
4899 | ||
4900 | if (pte_present(ptent)) | |
4901 | page = mc_handle_present_pte(vma, addr, ptent); | |
4902 | else if (is_swap_pte(ptent)) | |
48406ef8 | 4903 | page = mc_handle_swap_pte(vma, ptent, &ent); |
0661a336 | 4904 | else if (pte_none(ptent)) |
87946a72 | 4905 | page = mc_handle_file_pte(vma, addr, ptent, &ent); |
90254a65 DN |
4906 | |
4907 | if (!page && !ent.val) | |
8d32ff84 | 4908 | return ret; |
02491447 | 4909 | if (page) { |
02491447 | 4910 | /* |
0a31bc97 | 4911 | * Do only loose check w/o serialization. |
1306a85a | 4912 | * mem_cgroup_move_account() checks the page is valid or |
0a31bc97 | 4913 | * not under LRU exclusion. |
02491447 | 4914 | */ |
1306a85a | 4915 | if (page->mem_cgroup == mc.from) { |
02491447 | 4916 | ret = MC_TARGET_PAGE; |
df6ad698 JG |
4917 | if (is_device_private_page(page) || |
4918 | is_device_public_page(page)) | |
c733a828 | 4919 | ret = MC_TARGET_DEVICE; |
02491447 DN |
4920 | if (target) |
4921 | target->page = page; | |
4922 | } | |
4923 | if (!ret || !target) | |
4924 | put_page(page); | |
4925 | } | |
3e14a57b HY |
4926 | /* |
4927 | * There is a swap entry and a page doesn't exist or isn't charged. | |
4928 | * But we cannot move a tail-page in a THP. | |
4929 | */ | |
4930 | if (ent.val && !ret && (!page || !PageTransCompound(page)) && | |
34c00c31 | 4931 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
4932 | ret = MC_TARGET_SWAP; |
4933 | if (target) | |
4934 | target->ent = ent; | |
4ffef5fe | 4935 | } |
4ffef5fe DN |
4936 | return ret; |
4937 | } | |
4938 | ||
12724850 NH |
4939 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
4940 | /* | |
d6810d73 HY |
4941 | * We don't consider PMD mapped swapping or file mapped pages because THP does |
4942 | * not support them for now. | |
12724850 NH |
4943 | * Caller should make sure that pmd_trans_huge(pmd) is true. |
4944 | */ | |
4945 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
4946 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
4947 | { | |
4948 | struct page *page = NULL; | |
12724850 NH |
4949 | enum mc_target_type ret = MC_TARGET_NONE; |
4950 | ||
84c3fc4e ZY |
4951 | if (unlikely(is_swap_pmd(pmd))) { |
4952 | VM_BUG_ON(thp_migration_supported() && | |
4953 | !is_pmd_migration_entry(pmd)); | |
4954 | return ret; | |
4955 | } | |
12724850 | 4956 | page = pmd_page(pmd); |
309381fe | 4957 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
1dfab5ab | 4958 | if (!(mc.flags & MOVE_ANON)) |
12724850 | 4959 | return ret; |
1306a85a | 4960 | if (page->mem_cgroup == mc.from) { |
12724850 NH |
4961 | ret = MC_TARGET_PAGE; |
4962 | if (target) { | |
4963 | get_page(page); | |
4964 | target->page = page; | |
4965 | } | |
4966 | } | |
4967 | return ret; | |
4968 | } | |
4969 | #else | |
4970 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
4971 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
4972 | { | |
4973 | return MC_TARGET_NONE; | |
4974 | } | |
4975 | #endif | |
4976 | ||
4ffef5fe DN |
4977 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
4978 | unsigned long addr, unsigned long end, | |
4979 | struct mm_walk *walk) | |
4980 | { | |
26bcd64a | 4981 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
4982 | pte_t *pte; |
4983 | spinlock_t *ptl; | |
4984 | ||
b6ec57f4 KS |
4985 | ptl = pmd_trans_huge_lock(pmd, vma); |
4986 | if (ptl) { | |
c733a828 JG |
4987 | /* |
4988 | * Note their can not be MC_TARGET_DEVICE for now as we do not | |
4989 | * support transparent huge page with MEMORY_DEVICE_PUBLIC or | |
4990 | * MEMORY_DEVICE_PRIVATE but this might change. | |
4991 | */ | |
12724850 NH |
4992 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
4993 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 4994 | spin_unlock(ptl); |
1a5a9906 | 4995 | return 0; |
12724850 | 4996 | } |
03319327 | 4997 | |
45f83cef AA |
4998 | if (pmd_trans_unstable(pmd)) |
4999 | return 0; | |
4ffef5fe DN |
5000 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
5001 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 5002 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
5003 | mc.precharge++; /* increment precharge temporarily */ |
5004 | pte_unmap_unlock(pte - 1, ptl); | |
5005 | cond_resched(); | |
5006 | ||
7dc74be0 DN |
5007 | return 0; |
5008 | } | |
5009 | ||
4ffef5fe DN |
5010 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
5011 | { | |
5012 | unsigned long precharge; | |
4ffef5fe | 5013 | |
26bcd64a NH |
5014 | struct mm_walk mem_cgroup_count_precharge_walk = { |
5015 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
5016 | .mm = mm, | |
5017 | }; | |
dfe076b0 | 5018 | down_read(&mm->mmap_sem); |
0247f3f4 JM |
5019 | walk_page_range(0, mm->highest_vm_end, |
5020 | &mem_cgroup_count_precharge_walk); | |
dfe076b0 | 5021 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
5022 | |
5023 | precharge = mc.precharge; | |
5024 | mc.precharge = 0; | |
5025 | ||
5026 | return precharge; | |
5027 | } | |
5028 | ||
4ffef5fe DN |
5029 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
5030 | { | |
dfe076b0 DN |
5031 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
5032 | ||
5033 | VM_BUG_ON(mc.moving_task); | |
5034 | mc.moving_task = current; | |
5035 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
5036 | } |
5037 | ||
dfe076b0 DN |
5038 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
5039 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 5040 | { |
2bd9bb20 KH |
5041 | struct mem_cgroup *from = mc.from; |
5042 | struct mem_cgroup *to = mc.to; | |
5043 | ||
4ffef5fe | 5044 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d | 5045 | if (mc.precharge) { |
00501b53 | 5046 | cancel_charge(mc.to, mc.precharge); |
854ffa8d DN |
5047 | mc.precharge = 0; |
5048 | } | |
5049 | /* | |
5050 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
5051 | * we must uncharge here. | |
5052 | */ | |
5053 | if (mc.moved_charge) { | |
00501b53 | 5054 | cancel_charge(mc.from, mc.moved_charge); |
854ffa8d | 5055 | mc.moved_charge = 0; |
4ffef5fe | 5056 | } |
483c30b5 DN |
5057 | /* we must fixup refcnts and charges */ |
5058 | if (mc.moved_swap) { | |
483c30b5 | 5059 | /* uncharge swap account from the old cgroup */ |
ce00a967 | 5060 | if (!mem_cgroup_is_root(mc.from)) |
3e32cb2e | 5061 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
483c30b5 | 5062 | |
615d66c3 VD |
5063 | mem_cgroup_id_put_many(mc.from, mc.moved_swap); |
5064 | ||
05b84301 | 5065 | /* |
3e32cb2e JW |
5066 | * we charged both to->memory and to->memsw, so we |
5067 | * should uncharge to->memory. | |
05b84301 | 5068 | */ |
ce00a967 | 5069 | if (!mem_cgroup_is_root(mc.to)) |
3e32cb2e JW |
5070 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
5071 | ||
615d66c3 VD |
5072 | mem_cgroup_id_get_many(mc.to, mc.moved_swap); |
5073 | css_put_many(&mc.to->css, mc.moved_swap); | |
3e32cb2e | 5074 | |
483c30b5 DN |
5075 | mc.moved_swap = 0; |
5076 | } | |
dfe076b0 DN |
5077 | memcg_oom_recover(from); |
5078 | memcg_oom_recover(to); | |
5079 | wake_up_all(&mc.waitq); | |
5080 | } | |
5081 | ||
5082 | static void mem_cgroup_clear_mc(void) | |
5083 | { | |
264a0ae1 TH |
5084 | struct mm_struct *mm = mc.mm; |
5085 | ||
dfe076b0 DN |
5086 | /* |
5087 | * we must clear moving_task before waking up waiters at the end of | |
5088 | * task migration. | |
5089 | */ | |
5090 | mc.moving_task = NULL; | |
5091 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 5092 | spin_lock(&mc.lock); |
4ffef5fe DN |
5093 | mc.from = NULL; |
5094 | mc.to = NULL; | |
264a0ae1 | 5095 | mc.mm = NULL; |
2bd9bb20 | 5096 | spin_unlock(&mc.lock); |
264a0ae1 TH |
5097 | |
5098 | mmput(mm); | |
4ffef5fe DN |
5099 | } |
5100 | ||
1f7dd3e5 | 5101 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
7dc74be0 | 5102 | { |
1f7dd3e5 | 5103 | struct cgroup_subsys_state *css; |
eed67d75 | 5104 | struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */ |
9f2115f9 | 5105 | struct mem_cgroup *from; |
4530eddb | 5106 | struct task_struct *leader, *p; |
9f2115f9 | 5107 | struct mm_struct *mm; |
1dfab5ab | 5108 | unsigned long move_flags; |
9f2115f9 | 5109 | int ret = 0; |
7dc74be0 | 5110 | |
1f7dd3e5 TH |
5111 | /* charge immigration isn't supported on the default hierarchy */ |
5112 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
9f2115f9 TH |
5113 | return 0; |
5114 | ||
4530eddb TH |
5115 | /* |
5116 | * Multi-process migrations only happen on the default hierarchy | |
5117 | * where charge immigration is not used. Perform charge | |
5118 | * immigration if @tset contains a leader and whine if there are | |
5119 | * multiple. | |
5120 | */ | |
5121 | p = NULL; | |
1f7dd3e5 | 5122 | cgroup_taskset_for_each_leader(leader, css, tset) { |
4530eddb TH |
5123 | WARN_ON_ONCE(p); |
5124 | p = leader; | |
1f7dd3e5 | 5125 | memcg = mem_cgroup_from_css(css); |
4530eddb TH |
5126 | } |
5127 | if (!p) | |
5128 | return 0; | |
5129 | ||
1f7dd3e5 TH |
5130 | /* |
5131 | * We are now commited to this value whatever it is. Changes in this | |
5132 | * tunable will only affect upcoming migrations, not the current one. | |
5133 | * So we need to save it, and keep it going. | |
5134 | */ | |
5135 | move_flags = READ_ONCE(memcg->move_charge_at_immigrate); | |
5136 | if (!move_flags) | |
5137 | return 0; | |
5138 | ||
9f2115f9 TH |
5139 | from = mem_cgroup_from_task(p); |
5140 | ||
5141 | VM_BUG_ON(from == memcg); | |
5142 | ||
5143 | mm = get_task_mm(p); | |
5144 | if (!mm) | |
5145 | return 0; | |
5146 | /* We move charges only when we move a owner of the mm */ | |
5147 | if (mm->owner == p) { | |
5148 | VM_BUG_ON(mc.from); | |
5149 | VM_BUG_ON(mc.to); | |
5150 | VM_BUG_ON(mc.precharge); | |
5151 | VM_BUG_ON(mc.moved_charge); | |
5152 | VM_BUG_ON(mc.moved_swap); | |
5153 | ||
5154 | spin_lock(&mc.lock); | |
264a0ae1 | 5155 | mc.mm = mm; |
9f2115f9 TH |
5156 | mc.from = from; |
5157 | mc.to = memcg; | |
5158 | mc.flags = move_flags; | |
5159 | spin_unlock(&mc.lock); | |
5160 | /* We set mc.moving_task later */ | |
5161 | ||
5162 | ret = mem_cgroup_precharge_mc(mm); | |
5163 | if (ret) | |
5164 | mem_cgroup_clear_mc(); | |
264a0ae1 TH |
5165 | } else { |
5166 | mmput(mm); | |
7dc74be0 DN |
5167 | } |
5168 | return ret; | |
5169 | } | |
5170 | ||
1f7dd3e5 | 5171 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
7dc74be0 | 5172 | { |
4e2f245d JW |
5173 | if (mc.to) |
5174 | mem_cgroup_clear_mc(); | |
7dc74be0 DN |
5175 | } |
5176 | ||
4ffef5fe DN |
5177 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
5178 | unsigned long addr, unsigned long end, | |
5179 | struct mm_walk *walk) | |
7dc74be0 | 5180 | { |
4ffef5fe | 5181 | int ret = 0; |
26bcd64a | 5182 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
5183 | pte_t *pte; |
5184 | spinlock_t *ptl; | |
12724850 NH |
5185 | enum mc_target_type target_type; |
5186 | union mc_target target; | |
5187 | struct page *page; | |
4ffef5fe | 5188 | |
b6ec57f4 KS |
5189 | ptl = pmd_trans_huge_lock(pmd, vma); |
5190 | if (ptl) { | |
62ade86a | 5191 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 5192 | spin_unlock(ptl); |
12724850 NH |
5193 | return 0; |
5194 | } | |
5195 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
5196 | if (target_type == MC_TARGET_PAGE) { | |
5197 | page = target.page; | |
5198 | if (!isolate_lru_page(page)) { | |
f627c2f5 | 5199 | if (!mem_cgroup_move_account(page, true, |
1306a85a | 5200 | mc.from, mc.to)) { |
12724850 NH |
5201 | mc.precharge -= HPAGE_PMD_NR; |
5202 | mc.moved_charge += HPAGE_PMD_NR; | |
5203 | } | |
5204 | putback_lru_page(page); | |
5205 | } | |
5206 | put_page(page); | |
c733a828 JG |
5207 | } else if (target_type == MC_TARGET_DEVICE) { |
5208 | page = target.page; | |
5209 | if (!mem_cgroup_move_account(page, true, | |
5210 | mc.from, mc.to)) { | |
5211 | mc.precharge -= HPAGE_PMD_NR; | |
5212 | mc.moved_charge += HPAGE_PMD_NR; | |
5213 | } | |
5214 | put_page(page); | |
12724850 | 5215 | } |
bf929152 | 5216 | spin_unlock(ptl); |
1a5a9906 | 5217 | return 0; |
12724850 NH |
5218 | } |
5219 | ||
45f83cef AA |
5220 | if (pmd_trans_unstable(pmd)) |
5221 | return 0; | |
4ffef5fe DN |
5222 | retry: |
5223 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
5224 | for (; addr != end; addr += PAGE_SIZE) { | |
5225 | pte_t ptent = *(pte++); | |
c733a828 | 5226 | bool device = false; |
02491447 | 5227 | swp_entry_t ent; |
4ffef5fe DN |
5228 | |
5229 | if (!mc.precharge) | |
5230 | break; | |
5231 | ||
8d32ff84 | 5232 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
c733a828 JG |
5233 | case MC_TARGET_DEVICE: |
5234 | device = true; | |
5235 | /* fall through */ | |
4ffef5fe DN |
5236 | case MC_TARGET_PAGE: |
5237 | page = target.page; | |
53f9263b KS |
5238 | /* |
5239 | * We can have a part of the split pmd here. Moving it | |
5240 | * can be done but it would be too convoluted so simply | |
5241 | * ignore such a partial THP and keep it in original | |
5242 | * memcg. There should be somebody mapping the head. | |
5243 | */ | |
5244 | if (PageTransCompound(page)) | |
5245 | goto put; | |
c733a828 | 5246 | if (!device && isolate_lru_page(page)) |
4ffef5fe | 5247 | goto put; |
f627c2f5 KS |
5248 | if (!mem_cgroup_move_account(page, false, |
5249 | mc.from, mc.to)) { | |
4ffef5fe | 5250 | mc.precharge--; |
854ffa8d DN |
5251 | /* we uncharge from mc.from later. */ |
5252 | mc.moved_charge++; | |
4ffef5fe | 5253 | } |
c733a828 JG |
5254 | if (!device) |
5255 | putback_lru_page(page); | |
8d32ff84 | 5256 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
5257 | put_page(page); |
5258 | break; | |
02491447 DN |
5259 | case MC_TARGET_SWAP: |
5260 | ent = target.ent; | |
e91cbb42 | 5261 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 5262 | mc.precharge--; |
483c30b5 DN |
5263 | /* we fixup refcnts and charges later. */ |
5264 | mc.moved_swap++; | |
5265 | } | |
02491447 | 5266 | break; |
4ffef5fe DN |
5267 | default: |
5268 | break; | |
5269 | } | |
5270 | } | |
5271 | pte_unmap_unlock(pte - 1, ptl); | |
5272 | cond_resched(); | |
5273 | ||
5274 | if (addr != end) { | |
5275 | /* | |
5276 | * We have consumed all precharges we got in can_attach(). | |
5277 | * We try charge one by one, but don't do any additional | |
5278 | * charges to mc.to if we have failed in charge once in attach() | |
5279 | * phase. | |
5280 | */ | |
854ffa8d | 5281 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
5282 | if (!ret) |
5283 | goto retry; | |
5284 | } | |
5285 | ||
5286 | return ret; | |
5287 | } | |
5288 | ||
264a0ae1 | 5289 | static void mem_cgroup_move_charge(void) |
4ffef5fe | 5290 | { |
26bcd64a NH |
5291 | struct mm_walk mem_cgroup_move_charge_walk = { |
5292 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
264a0ae1 | 5293 | .mm = mc.mm, |
26bcd64a | 5294 | }; |
4ffef5fe DN |
5295 | |
5296 | lru_add_drain_all(); | |
312722cb | 5297 | /* |
81f8c3a4 JW |
5298 | * Signal lock_page_memcg() to take the memcg's move_lock |
5299 | * while we're moving its pages to another memcg. Then wait | |
5300 | * for already started RCU-only updates to finish. | |
312722cb JW |
5301 | */ |
5302 | atomic_inc(&mc.from->moving_account); | |
5303 | synchronize_rcu(); | |
dfe076b0 | 5304 | retry: |
264a0ae1 | 5305 | if (unlikely(!down_read_trylock(&mc.mm->mmap_sem))) { |
dfe076b0 DN |
5306 | /* |
5307 | * Someone who are holding the mmap_sem might be waiting in | |
5308 | * waitq. So we cancel all extra charges, wake up all waiters, | |
5309 | * and retry. Because we cancel precharges, we might not be able | |
5310 | * to move enough charges, but moving charge is a best-effort | |
5311 | * feature anyway, so it wouldn't be a big problem. | |
5312 | */ | |
5313 | __mem_cgroup_clear_mc(); | |
5314 | cond_resched(); | |
5315 | goto retry; | |
5316 | } | |
26bcd64a NH |
5317 | /* |
5318 | * When we have consumed all precharges and failed in doing | |
5319 | * additional charge, the page walk just aborts. | |
5320 | */ | |
0247f3f4 JM |
5321 | walk_page_range(0, mc.mm->highest_vm_end, &mem_cgroup_move_charge_walk); |
5322 | ||
264a0ae1 | 5323 | up_read(&mc.mm->mmap_sem); |
312722cb | 5324 | atomic_dec(&mc.from->moving_account); |
7dc74be0 DN |
5325 | } |
5326 | ||
264a0ae1 | 5327 | static void mem_cgroup_move_task(void) |
67e465a7 | 5328 | { |
264a0ae1 TH |
5329 | if (mc.to) { |
5330 | mem_cgroup_move_charge(); | |
a433658c | 5331 | mem_cgroup_clear_mc(); |
264a0ae1 | 5332 | } |
67e465a7 | 5333 | } |
5cfb80a7 | 5334 | #else /* !CONFIG_MMU */ |
1f7dd3e5 | 5335 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
5336 | { |
5337 | return 0; | |
5338 | } | |
1f7dd3e5 | 5339 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
5340 | { |
5341 | } | |
264a0ae1 | 5342 | static void mem_cgroup_move_task(void) |
5cfb80a7 DN |
5343 | { |
5344 | } | |
5345 | #endif | |
67e465a7 | 5346 | |
f00baae7 TH |
5347 | /* |
5348 | * Cgroup retains root cgroups across [un]mount cycles making it necessary | |
aa6ec29b TH |
5349 | * to verify whether we're attached to the default hierarchy on each mount |
5350 | * attempt. | |
f00baae7 | 5351 | */ |
eb95419b | 5352 | static void mem_cgroup_bind(struct cgroup_subsys_state *root_css) |
f00baae7 TH |
5353 | { |
5354 | /* | |
aa6ec29b | 5355 | * use_hierarchy is forced on the default hierarchy. cgroup core |
f00baae7 TH |
5356 | * guarantees that @root doesn't have any children, so turning it |
5357 | * on for the root memcg is enough. | |
5358 | */ | |
9e10a130 | 5359 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
7feee590 VD |
5360 | root_mem_cgroup->use_hierarchy = true; |
5361 | else | |
5362 | root_mem_cgroup->use_hierarchy = false; | |
f00baae7 TH |
5363 | } |
5364 | ||
677dc973 CD |
5365 | static int seq_puts_memcg_tunable(struct seq_file *m, unsigned long value) |
5366 | { | |
5367 | if (value == PAGE_COUNTER_MAX) | |
5368 | seq_puts(m, "max\n"); | |
5369 | else | |
5370 | seq_printf(m, "%llu\n", (u64)value * PAGE_SIZE); | |
5371 | ||
5372 | return 0; | |
5373 | } | |
5374 | ||
241994ed JW |
5375 | static u64 memory_current_read(struct cgroup_subsys_state *css, |
5376 | struct cftype *cft) | |
5377 | { | |
f5fc3c5d JW |
5378 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5379 | ||
5380 | return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE; | |
241994ed JW |
5381 | } |
5382 | ||
bf8d5d52 RG |
5383 | static int memory_min_show(struct seq_file *m, void *v) |
5384 | { | |
677dc973 CD |
5385 | return seq_puts_memcg_tunable(m, |
5386 | READ_ONCE(mem_cgroup_from_seq(m)->memory.min)); | |
bf8d5d52 RG |
5387 | } |
5388 | ||
5389 | static ssize_t memory_min_write(struct kernfs_open_file *of, | |
5390 | char *buf, size_t nbytes, loff_t off) | |
5391 | { | |
5392 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5393 | unsigned long min; | |
5394 | int err; | |
5395 | ||
5396 | buf = strstrip(buf); | |
5397 | err = page_counter_memparse(buf, "max", &min); | |
5398 | if (err) | |
5399 | return err; | |
5400 | ||
5401 | page_counter_set_min(&memcg->memory, min); | |
5402 | ||
5403 | return nbytes; | |
5404 | } | |
5405 | ||
241994ed JW |
5406 | static int memory_low_show(struct seq_file *m, void *v) |
5407 | { | |
677dc973 CD |
5408 | return seq_puts_memcg_tunable(m, |
5409 | READ_ONCE(mem_cgroup_from_seq(m)->memory.low)); | |
241994ed JW |
5410 | } |
5411 | ||
5412 | static ssize_t memory_low_write(struct kernfs_open_file *of, | |
5413 | char *buf, size_t nbytes, loff_t off) | |
5414 | { | |
5415 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5416 | unsigned long low; | |
5417 | int err; | |
5418 | ||
5419 | buf = strstrip(buf); | |
d2973697 | 5420 | err = page_counter_memparse(buf, "max", &low); |
241994ed JW |
5421 | if (err) |
5422 | return err; | |
5423 | ||
23067153 | 5424 | page_counter_set_low(&memcg->memory, low); |
241994ed JW |
5425 | |
5426 | return nbytes; | |
5427 | } | |
5428 | ||
5429 | static int memory_high_show(struct seq_file *m, void *v) | |
5430 | { | |
677dc973 | 5431 | return seq_puts_memcg_tunable(m, READ_ONCE(mem_cgroup_from_seq(m)->high)); |
241994ed JW |
5432 | } |
5433 | ||
5434 | static ssize_t memory_high_write(struct kernfs_open_file *of, | |
5435 | char *buf, size_t nbytes, loff_t off) | |
5436 | { | |
5437 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
588083bb | 5438 | unsigned long nr_pages; |
241994ed JW |
5439 | unsigned long high; |
5440 | int err; | |
5441 | ||
5442 | buf = strstrip(buf); | |
d2973697 | 5443 | err = page_counter_memparse(buf, "max", &high); |
241994ed JW |
5444 | if (err) |
5445 | return err; | |
5446 | ||
5447 | memcg->high = high; | |
5448 | ||
588083bb JW |
5449 | nr_pages = page_counter_read(&memcg->memory); |
5450 | if (nr_pages > high) | |
5451 | try_to_free_mem_cgroup_pages(memcg, nr_pages - high, | |
5452 | GFP_KERNEL, true); | |
5453 | ||
2529bb3a | 5454 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
5455 | return nbytes; |
5456 | } | |
5457 | ||
5458 | static int memory_max_show(struct seq_file *m, void *v) | |
5459 | { | |
677dc973 CD |
5460 | return seq_puts_memcg_tunable(m, |
5461 | READ_ONCE(mem_cgroup_from_seq(m)->memory.max)); | |
241994ed JW |
5462 | } |
5463 | ||
5464 | static ssize_t memory_max_write(struct kernfs_open_file *of, | |
5465 | char *buf, size_t nbytes, loff_t off) | |
5466 | { | |
5467 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
b6e6edcf JW |
5468 | unsigned int nr_reclaims = MEM_CGROUP_RECLAIM_RETRIES; |
5469 | bool drained = false; | |
241994ed JW |
5470 | unsigned long max; |
5471 | int err; | |
5472 | ||
5473 | buf = strstrip(buf); | |
d2973697 | 5474 | err = page_counter_memparse(buf, "max", &max); |
241994ed JW |
5475 | if (err) |
5476 | return err; | |
5477 | ||
bbec2e15 | 5478 | xchg(&memcg->memory.max, max); |
b6e6edcf JW |
5479 | |
5480 | for (;;) { | |
5481 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
5482 | ||
5483 | if (nr_pages <= max) | |
5484 | break; | |
5485 | ||
5486 | if (signal_pending(current)) { | |
5487 | err = -EINTR; | |
5488 | break; | |
5489 | } | |
5490 | ||
5491 | if (!drained) { | |
5492 | drain_all_stock(memcg); | |
5493 | drained = true; | |
5494 | continue; | |
5495 | } | |
5496 | ||
5497 | if (nr_reclaims) { | |
5498 | if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max, | |
5499 | GFP_KERNEL, true)) | |
5500 | nr_reclaims--; | |
5501 | continue; | |
5502 | } | |
5503 | ||
e27be240 | 5504 | memcg_memory_event(memcg, MEMCG_OOM); |
b6e6edcf JW |
5505 | if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0)) |
5506 | break; | |
5507 | } | |
241994ed | 5508 | |
2529bb3a | 5509 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
5510 | return nbytes; |
5511 | } | |
5512 | ||
5513 | static int memory_events_show(struct seq_file *m, void *v) | |
5514 | { | |
aa9694bb | 5515 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
241994ed | 5516 | |
e27be240 JW |
5517 | seq_printf(m, "low %lu\n", |
5518 | atomic_long_read(&memcg->memory_events[MEMCG_LOW])); | |
5519 | seq_printf(m, "high %lu\n", | |
5520 | atomic_long_read(&memcg->memory_events[MEMCG_HIGH])); | |
5521 | seq_printf(m, "max %lu\n", | |
5522 | atomic_long_read(&memcg->memory_events[MEMCG_MAX])); | |
5523 | seq_printf(m, "oom %lu\n", | |
5524 | atomic_long_read(&memcg->memory_events[MEMCG_OOM])); | |
fe6bdfc8 RG |
5525 | seq_printf(m, "oom_kill %lu\n", |
5526 | atomic_long_read(&memcg->memory_events[MEMCG_OOM_KILL])); | |
241994ed JW |
5527 | |
5528 | return 0; | |
5529 | } | |
5530 | ||
587d9f72 JW |
5531 | static int memory_stat_show(struct seq_file *m, void *v) |
5532 | { | |
aa9694bb | 5533 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
8de7ecc6 | 5534 | struct accumulated_stats acc; |
587d9f72 JW |
5535 | int i; |
5536 | ||
5537 | /* | |
5538 | * Provide statistics on the state of the memory subsystem as | |
5539 | * well as cumulative event counters that show past behavior. | |
5540 | * | |
5541 | * This list is ordered following a combination of these gradients: | |
5542 | * 1) generic big picture -> specifics and details | |
5543 | * 2) reflecting userspace activity -> reflecting kernel heuristics | |
5544 | * | |
5545 | * Current memory state: | |
5546 | */ | |
5547 | ||
8de7ecc6 SB |
5548 | memset(&acc, 0, sizeof(acc)); |
5549 | acc.stats_size = MEMCG_NR_STAT; | |
5550 | acc.events_size = NR_VM_EVENT_ITEMS; | |
5551 | accumulate_memcg_tree(memcg, &acc); | |
72b54e73 | 5552 | |
587d9f72 | 5553 | seq_printf(m, "anon %llu\n", |
8de7ecc6 | 5554 | (u64)acc.stat[MEMCG_RSS] * PAGE_SIZE); |
587d9f72 | 5555 | seq_printf(m, "file %llu\n", |
8de7ecc6 | 5556 | (u64)acc.stat[MEMCG_CACHE] * PAGE_SIZE); |
12580e4b | 5557 | seq_printf(m, "kernel_stack %llu\n", |
8de7ecc6 | 5558 | (u64)acc.stat[MEMCG_KERNEL_STACK_KB] * 1024); |
27ee57c9 | 5559 | seq_printf(m, "slab %llu\n", |
8de7ecc6 SB |
5560 | (u64)(acc.stat[NR_SLAB_RECLAIMABLE] + |
5561 | acc.stat[NR_SLAB_UNRECLAIMABLE]) * PAGE_SIZE); | |
b2807f07 | 5562 | seq_printf(m, "sock %llu\n", |
8de7ecc6 | 5563 | (u64)acc.stat[MEMCG_SOCK] * PAGE_SIZE); |
587d9f72 | 5564 | |
9a4caf1e | 5565 | seq_printf(m, "shmem %llu\n", |
8de7ecc6 | 5566 | (u64)acc.stat[NR_SHMEM] * PAGE_SIZE); |
587d9f72 | 5567 | seq_printf(m, "file_mapped %llu\n", |
8de7ecc6 | 5568 | (u64)acc.stat[NR_FILE_MAPPED] * PAGE_SIZE); |
587d9f72 | 5569 | seq_printf(m, "file_dirty %llu\n", |
8de7ecc6 | 5570 | (u64)acc.stat[NR_FILE_DIRTY] * PAGE_SIZE); |
587d9f72 | 5571 | seq_printf(m, "file_writeback %llu\n", |
8de7ecc6 | 5572 | (u64)acc.stat[NR_WRITEBACK] * PAGE_SIZE); |
587d9f72 | 5573 | |
8de7ecc6 SB |
5574 | for (i = 0; i < NR_LRU_LISTS; i++) |
5575 | seq_printf(m, "%s %llu\n", mem_cgroup_lru_names[i], | |
5576 | (u64)acc.lru_pages[i] * PAGE_SIZE); | |
587d9f72 | 5577 | |
27ee57c9 | 5578 | seq_printf(m, "slab_reclaimable %llu\n", |
8de7ecc6 | 5579 | (u64)acc.stat[NR_SLAB_RECLAIMABLE] * PAGE_SIZE); |
27ee57c9 | 5580 | seq_printf(m, "slab_unreclaimable %llu\n", |
8de7ecc6 | 5581 | (u64)acc.stat[NR_SLAB_UNRECLAIMABLE] * PAGE_SIZE); |
27ee57c9 | 5582 | |
587d9f72 JW |
5583 | /* Accumulated memory events */ |
5584 | ||
8de7ecc6 SB |
5585 | seq_printf(m, "pgfault %lu\n", acc.events[PGFAULT]); |
5586 | seq_printf(m, "pgmajfault %lu\n", acc.events[PGMAJFAULT]); | |
587d9f72 | 5587 | |
e9b257ed JW |
5588 | seq_printf(m, "workingset_refault %lu\n", |
5589 | acc.stat[WORKINGSET_REFAULT]); | |
5590 | seq_printf(m, "workingset_activate %lu\n", | |
5591 | acc.stat[WORKINGSET_ACTIVATE]); | |
5592 | seq_printf(m, "workingset_nodereclaim %lu\n", | |
5593 | acc.stat[WORKINGSET_NODERECLAIM]); | |
5594 | ||
8de7ecc6 SB |
5595 | seq_printf(m, "pgrefill %lu\n", acc.events[PGREFILL]); |
5596 | seq_printf(m, "pgscan %lu\n", acc.events[PGSCAN_KSWAPD] + | |
5597 | acc.events[PGSCAN_DIRECT]); | |
5598 | seq_printf(m, "pgsteal %lu\n", acc.events[PGSTEAL_KSWAPD] + | |
5599 | acc.events[PGSTEAL_DIRECT]); | |
5600 | seq_printf(m, "pgactivate %lu\n", acc.events[PGACTIVATE]); | |
5601 | seq_printf(m, "pgdeactivate %lu\n", acc.events[PGDEACTIVATE]); | |
5602 | seq_printf(m, "pglazyfree %lu\n", acc.events[PGLAZYFREE]); | |
5603 | seq_printf(m, "pglazyfreed %lu\n", acc.events[PGLAZYFREED]); | |
2262185c | 5604 | |
587d9f72 JW |
5605 | return 0; |
5606 | } | |
5607 | ||
3d8b38eb RG |
5608 | static int memory_oom_group_show(struct seq_file *m, void *v) |
5609 | { | |
aa9694bb | 5610 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
3d8b38eb RG |
5611 | |
5612 | seq_printf(m, "%d\n", memcg->oom_group); | |
5613 | ||
5614 | return 0; | |
5615 | } | |
5616 | ||
5617 | static ssize_t memory_oom_group_write(struct kernfs_open_file *of, | |
5618 | char *buf, size_t nbytes, loff_t off) | |
5619 | { | |
5620 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5621 | int ret, oom_group; | |
5622 | ||
5623 | buf = strstrip(buf); | |
5624 | if (!buf) | |
5625 | return -EINVAL; | |
5626 | ||
5627 | ret = kstrtoint(buf, 0, &oom_group); | |
5628 | if (ret) | |
5629 | return ret; | |
5630 | ||
5631 | if (oom_group != 0 && oom_group != 1) | |
5632 | return -EINVAL; | |
5633 | ||
5634 | memcg->oom_group = oom_group; | |
5635 | ||
5636 | return nbytes; | |
5637 | } | |
5638 | ||
241994ed JW |
5639 | static struct cftype memory_files[] = { |
5640 | { | |
5641 | .name = "current", | |
f5fc3c5d | 5642 | .flags = CFTYPE_NOT_ON_ROOT, |
241994ed JW |
5643 | .read_u64 = memory_current_read, |
5644 | }, | |
bf8d5d52 RG |
5645 | { |
5646 | .name = "min", | |
5647 | .flags = CFTYPE_NOT_ON_ROOT, | |
5648 | .seq_show = memory_min_show, | |
5649 | .write = memory_min_write, | |
5650 | }, | |
241994ed JW |
5651 | { |
5652 | .name = "low", | |
5653 | .flags = CFTYPE_NOT_ON_ROOT, | |
5654 | .seq_show = memory_low_show, | |
5655 | .write = memory_low_write, | |
5656 | }, | |
5657 | { | |
5658 | .name = "high", | |
5659 | .flags = CFTYPE_NOT_ON_ROOT, | |
5660 | .seq_show = memory_high_show, | |
5661 | .write = memory_high_write, | |
5662 | }, | |
5663 | { | |
5664 | .name = "max", | |
5665 | .flags = CFTYPE_NOT_ON_ROOT, | |
5666 | .seq_show = memory_max_show, | |
5667 | .write = memory_max_write, | |
5668 | }, | |
5669 | { | |
5670 | .name = "events", | |
5671 | .flags = CFTYPE_NOT_ON_ROOT, | |
472912a2 | 5672 | .file_offset = offsetof(struct mem_cgroup, events_file), |
241994ed JW |
5673 | .seq_show = memory_events_show, |
5674 | }, | |
587d9f72 JW |
5675 | { |
5676 | .name = "stat", | |
5677 | .flags = CFTYPE_NOT_ON_ROOT, | |
5678 | .seq_show = memory_stat_show, | |
5679 | }, | |
3d8b38eb RG |
5680 | { |
5681 | .name = "oom.group", | |
5682 | .flags = CFTYPE_NOT_ON_ROOT | CFTYPE_NS_DELEGATABLE, | |
5683 | .seq_show = memory_oom_group_show, | |
5684 | .write = memory_oom_group_write, | |
5685 | }, | |
241994ed JW |
5686 | { } /* terminate */ |
5687 | }; | |
5688 | ||
073219e9 | 5689 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 5690 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 5691 | .css_online = mem_cgroup_css_online, |
92fb9748 | 5692 | .css_offline = mem_cgroup_css_offline, |
6df38689 | 5693 | .css_released = mem_cgroup_css_released, |
92fb9748 | 5694 | .css_free = mem_cgroup_css_free, |
1ced953b | 5695 | .css_reset = mem_cgroup_css_reset, |
7dc74be0 DN |
5696 | .can_attach = mem_cgroup_can_attach, |
5697 | .cancel_attach = mem_cgroup_cancel_attach, | |
264a0ae1 | 5698 | .post_attach = mem_cgroup_move_task, |
f00baae7 | 5699 | .bind = mem_cgroup_bind, |
241994ed JW |
5700 | .dfl_cftypes = memory_files, |
5701 | .legacy_cftypes = mem_cgroup_legacy_files, | |
6d12e2d8 | 5702 | .early_init = 0, |
8cdea7c0 | 5703 | }; |
c077719b | 5704 | |
241994ed | 5705 | /** |
bf8d5d52 | 5706 | * mem_cgroup_protected - check if memory consumption is in the normal range |
34c81057 | 5707 | * @root: the top ancestor of the sub-tree being checked |
241994ed JW |
5708 | * @memcg: the memory cgroup to check |
5709 | * | |
23067153 RG |
5710 | * WARNING: This function is not stateless! It can only be used as part |
5711 | * of a top-down tree iteration, not for isolated queries. | |
34c81057 | 5712 | * |
bf8d5d52 RG |
5713 | * Returns one of the following: |
5714 | * MEMCG_PROT_NONE: cgroup memory is not protected | |
5715 | * MEMCG_PROT_LOW: cgroup memory is protected as long there is | |
5716 | * an unprotected supply of reclaimable memory from other cgroups. | |
5717 | * MEMCG_PROT_MIN: cgroup memory is protected | |
34c81057 | 5718 | * |
bf8d5d52 | 5719 | * @root is exclusive; it is never protected when looked at directly |
34c81057 | 5720 | * |
bf8d5d52 RG |
5721 | * To provide a proper hierarchical behavior, effective memory.min/low values |
5722 | * are used. Below is the description of how effective memory.low is calculated. | |
5723 | * Effective memory.min values is calculated in the same way. | |
34c81057 | 5724 | * |
23067153 RG |
5725 | * Effective memory.low is always equal or less than the original memory.low. |
5726 | * If there is no memory.low overcommittment (which is always true for | |
5727 | * top-level memory cgroups), these two values are equal. | |
5728 | * Otherwise, it's a part of parent's effective memory.low, | |
5729 | * calculated as a cgroup's memory.low usage divided by sum of sibling's | |
5730 | * memory.low usages, where memory.low usage is the size of actually | |
5731 | * protected memory. | |
34c81057 | 5732 | * |
23067153 RG |
5733 | * low_usage |
5734 | * elow = min( memory.low, parent->elow * ------------------ ), | |
5735 | * siblings_low_usage | |
34c81057 | 5736 | * |
23067153 RG |
5737 | * | memory.current, if memory.current < memory.low |
5738 | * low_usage = | | |
5739 | | 0, otherwise. | |
34c81057 | 5740 | * |
23067153 RG |
5741 | * |
5742 | * Such definition of the effective memory.low provides the expected | |
5743 | * hierarchical behavior: parent's memory.low value is limiting | |
5744 | * children, unprotected memory is reclaimed first and cgroups, | |
5745 | * which are not using their guarantee do not affect actual memory | |
5746 | * distribution. | |
5747 | * | |
5748 | * For example, if there are memcgs A, A/B, A/C, A/D and A/E: | |
5749 | * | |
5750 | * A A/memory.low = 2G, A/memory.current = 6G | |
5751 | * //\\ | |
5752 | * BC DE B/memory.low = 3G B/memory.current = 2G | |
5753 | * C/memory.low = 1G C/memory.current = 2G | |
5754 | * D/memory.low = 0 D/memory.current = 2G | |
5755 | * E/memory.low = 10G E/memory.current = 0 | |
5756 | * | |
5757 | * and the memory pressure is applied, the following memory distribution | |
5758 | * is expected (approximately): | |
5759 | * | |
5760 | * A/memory.current = 2G | |
5761 | * | |
5762 | * B/memory.current = 1.3G | |
5763 | * C/memory.current = 0.6G | |
5764 | * D/memory.current = 0 | |
5765 | * E/memory.current = 0 | |
5766 | * | |
5767 | * These calculations require constant tracking of the actual low usages | |
bf8d5d52 RG |
5768 | * (see propagate_protected_usage()), as well as recursive calculation of |
5769 | * effective memory.low values. But as we do call mem_cgroup_protected() | |
23067153 RG |
5770 | * path for each memory cgroup top-down from the reclaim, |
5771 | * it's possible to optimize this part, and save calculated elow | |
5772 | * for next usage. This part is intentionally racy, but it's ok, | |
5773 | * as memory.low is a best-effort mechanism. | |
241994ed | 5774 | */ |
bf8d5d52 RG |
5775 | enum mem_cgroup_protection mem_cgroup_protected(struct mem_cgroup *root, |
5776 | struct mem_cgroup *memcg) | |
241994ed | 5777 | { |
23067153 | 5778 | struct mem_cgroup *parent; |
bf8d5d52 RG |
5779 | unsigned long emin, parent_emin; |
5780 | unsigned long elow, parent_elow; | |
5781 | unsigned long usage; | |
23067153 | 5782 | |
241994ed | 5783 | if (mem_cgroup_disabled()) |
bf8d5d52 | 5784 | return MEMCG_PROT_NONE; |
241994ed | 5785 | |
34c81057 SC |
5786 | if (!root) |
5787 | root = root_mem_cgroup; | |
5788 | if (memcg == root) | |
bf8d5d52 | 5789 | return MEMCG_PROT_NONE; |
241994ed | 5790 | |
23067153 | 5791 | usage = page_counter_read(&memcg->memory); |
bf8d5d52 RG |
5792 | if (!usage) |
5793 | return MEMCG_PROT_NONE; | |
5794 | ||
5795 | emin = memcg->memory.min; | |
5796 | elow = memcg->memory.low; | |
34c81057 | 5797 | |
bf8d5d52 | 5798 | parent = parent_mem_cgroup(memcg); |
df2a4196 RG |
5799 | /* No parent means a non-hierarchical mode on v1 memcg */ |
5800 | if (!parent) | |
5801 | return MEMCG_PROT_NONE; | |
5802 | ||
23067153 RG |
5803 | if (parent == root) |
5804 | goto exit; | |
5805 | ||
bf8d5d52 RG |
5806 | parent_emin = READ_ONCE(parent->memory.emin); |
5807 | emin = min(emin, parent_emin); | |
5808 | if (emin && parent_emin) { | |
5809 | unsigned long min_usage, siblings_min_usage; | |
5810 | ||
5811 | min_usage = min(usage, memcg->memory.min); | |
5812 | siblings_min_usage = atomic_long_read( | |
5813 | &parent->memory.children_min_usage); | |
5814 | ||
5815 | if (min_usage && siblings_min_usage) | |
5816 | emin = min(emin, parent_emin * min_usage / | |
5817 | siblings_min_usage); | |
5818 | } | |
5819 | ||
23067153 RG |
5820 | parent_elow = READ_ONCE(parent->memory.elow); |
5821 | elow = min(elow, parent_elow); | |
bf8d5d52 RG |
5822 | if (elow && parent_elow) { |
5823 | unsigned long low_usage, siblings_low_usage; | |
23067153 | 5824 | |
bf8d5d52 RG |
5825 | low_usage = min(usage, memcg->memory.low); |
5826 | siblings_low_usage = atomic_long_read( | |
5827 | &parent->memory.children_low_usage); | |
23067153 | 5828 | |
bf8d5d52 RG |
5829 | if (low_usage && siblings_low_usage) |
5830 | elow = min(elow, parent_elow * low_usage / | |
5831 | siblings_low_usage); | |
5832 | } | |
23067153 | 5833 | |
23067153 | 5834 | exit: |
bf8d5d52 | 5835 | memcg->memory.emin = emin; |
23067153 | 5836 | memcg->memory.elow = elow; |
bf8d5d52 RG |
5837 | |
5838 | if (usage <= emin) | |
5839 | return MEMCG_PROT_MIN; | |
5840 | else if (usage <= elow) | |
5841 | return MEMCG_PROT_LOW; | |
5842 | else | |
5843 | return MEMCG_PROT_NONE; | |
241994ed JW |
5844 | } |
5845 | ||
00501b53 JW |
5846 | /** |
5847 | * mem_cgroup_try_charge - try charging a page | |
5848 | * @page: page to charge | |
5849 | * @mm: mm context of the victim | |
5850 | * @gfp_mask: reclaim mode | |
5851 | * @memcgp: charged memcg return | |
25843c2b | 5852 | * @compound: charge the page as compound or small page |
00501b53 JW |
5853 | * |
5854 | * Try to charge @page to the memcg that @mm belongs to, reclaiming | |
5855 | * pages according to @gfp_mask if necessary. | |
5856 | * | |
5857 | * Returns 0 on success, with *@memcgp pointing to the charged memcg. | |
5858 | * Otherwise, an error code is returned. | |
5859 | * | |
5860 | * After page->mapping has been set up, the caller must finalize the | |
5861 | * charge with mem_cgroup_commit_charge(). Or abort the transaction | |
5862 | * with mem_cgroup_cancel_charge() in case page instantiation fails. | |
5863 | */ | |
5864 | int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm, | |
f627c2f5 KS |
5865 | gfp_t gfp_mask, struct mem_cgroup **memcgp, |
5866 | bool compound) | |
00501b53 JW |
5867 | { |
5868 | struct mem_cgroup *memcg = NULL; | |
f627c2f5 | 5869 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
00501b53 JW |
5870 | int ret = 0; |
5871 | ||
5872 | if (mem_cgroup_disabled()) | |
5873 | goto out; | |
5874 | ||
5875 | if (PageSwapCache(page)) { | |
00501b53 JW |
5876 | /* |
5877 | * Every swap fault against a single page tries to charge the | |
5878 | * page, bail as early as possible. shmem_unuse() encounters | |
5879 | * already charged pages, too. The USED bit is protected by | |
5880 | * the page lock, which serializes swap cache removal, which | |
5881 | * in turn serializes uncharging. | |
5882 | */ | |
e993d905 | 5883 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
abe2895b | 5884 | if (compound_head(page)->mem_cgroup) |
00501b53 | 5885 | goto out; |
e993d905 | 5886 | |
37e84351 | 5887 | if (do_swap_account) { |
e993d905 VD |
5888 | swp_entry_t ent = { .val = page_private(page), }; |
5889 | unsigned short id = lookup_swap_cgroup_id(ent); | |
5890 | ||
5891 | rcu_read_lock(); | |
5892 | memcg = mem_cgroup_from_id(id); | |
5893 | if (memcg && !css_tryget_online(&memcg->css)) | |
5894 | memcg = NULL; | |
5895 | rcu_read_unlock(); | |
5896 | } | |
00501b53 JW |
5897 | } |
5898 | ||
00501b53 JW |
5899 | if (!memcg) |
5900 | memcg = get_mem_cgroup_from_mm(mm); | |
5901 | ||
5902 | ret = try_charge(memcg, gfp_mask, nr_pages); | |
5903 | ||
5904 | css_put(&memcg->css); | |
00501b53 JW |
5905 | out: |
5906 | *memcgp = memcg; | |
5907 | return ret; | |
5908 | } | |
5909 | ||
2cf85583 TH |
5910 | int mem_cgroup_try_charge_delay(struct page *page, struct mm_struct *mm, |
5911 | gfp_t gfp_mask, struct mem_cgroup **memcgp, | |
5912 | bool compound) | |
5913 | { | |
5914 | struct mem_cgroup *memcg; | |
5915 | int ret; | |
5916 | ||
5917 | ret = mem_cgroup_try_charge(page, mm, gfp_mask, memcgp, compound); | |
5918 | memcg = *memcgp; | |
5919 | mem_cgroup_throttle_swaprate(memcg, page_to_nid(page), gfp_mask); | |
5920 | return ret; | |
5921 | } | |
5922 | ||
00501b53 JW |
5923 | /** |
5924 | * mem_cgroup_commit_charge - commit a page charge | |
5925 | * @page: page to charge | |
5926 | * @memcg: memcg to charge the page to | |
5927 | * @lrucare: page might be on LRU already | |
25843c2b | 5928 | * @compound: charge the page as compound or small page |
00501b53 JW |
5929 | * |
5930 | * Finalize a charge transaction started by mem_cgroup_try_charge(), | |
5931 | * after page->mapping has been set up. This must happen atomically | |
5932 | * as part of the page instantiation, i.e. under the page table lock | |
5933 | * for anonymous pages, under the page lock for page and swap cache. | |
5934 | * | |
5935 | * In addition, the page must not be on the LRU during the commit, to | |
5936 | * prevent racing with task migration. If it might be, use @lrucare. | |
5937 | * | |
5938 | * Use mem_cgroup_cancel_charge() to cancel the transaction instead. | |
5939 | */ | |
5940 | void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
f627c2f5 | 5941 | bool lrucare, bool compound) |
00501b53 | 5942 | { |
f627c2f5 | 5943 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
00501b53 JW |
5944 | |
5945 | VM_BUG_ON_PAGE(!page->mapping, page); | |
5946 | VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page); | |
5947 | ||
5948 | if (mem_cgroup_disabled()) | |
5949 | return; | |
5950 | /* | |
5951 | * Swap faults will attempt to charge the same page multiple | |
5952 | * times. But reuse_swap_page() might have removed the page | |
5953 | * from swapcache already, so we can't check PageSwapCache(). | |
5954 | */ | |
5955 | if (!memcg) | |
5956 | return; | |
5957 | ||
6abb5a86 JW |
5958 | commit_charge(page, memcg, lrucare); |
5959 | ||
6abb5a86 | 5960 | local_irq_disable(); |
f627c2f5 | 5961 | mem_cgroup_charge_statistics(memcg, page, compound, nr_pages); |
6abb5a86 JW |
5962 | memcg_check_events(memcg, page); |
5963 | local_irq_enable(); | |
00501b53 | 5964 | |
7941d214 | 5965 | if (do_memsw_account() && PageSwapCache(page)) { |
00501b53 JW |
5966 | swp_entry_t entry = { .val = page_private(page) }; |
5967 | /* | |
5968 | * The swap entry might not get freed for a long time, | |
5969 | * let's not wait for it. The page already received a | |
5970 | * memory+swap charge, drop the swap entry duplicate. | |
5971 | */ | |
38d8b4e6 | 5972 | mem_cgroup_uncharge_swap(entry, nr_pages); |
00501b53 JW |
5973 | } |
5974 | } | |
5975 | ||
5976 | /** | |
5977 | * mem_cgroup_cancel_charge - cancel a page charge | |
5978 | * @page: page to charge | |
5979 | * @memcg: memcg to charge the page to | |
25843c2b | 5980 | * @compound: charge the page as compound or small page |
00501b53 JW |
5981 | * |
5982 | * Cancel a charge transaction started by mem_cgroup_try_charge(). | |
5983 | */ | |
f627c2f5 KS |
5984 | void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg, |
5985 | bool compound) | |
00501b53 | 5986 | { |
f627c2f5 | 5987 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
00501b53 JW |
5988 | |
5989 | if (mem_cgroup_disabled()) | |
5990 | return; | |
5991 | /* | |
5992 | * Swap faults will attempt to charge the same page multiple | |
5993 | * times. But reuse_swap_page() might have removed the page | |
5994 | * from swapcache already, so we can't check PageSwapCache(). | |
5995 | */ | |
5996 | if (!memcg) | |
5997 | return; | |
5998 | ||
00501b53 JW |
5999 | cancel_charge(memcg, nr_pages); |
6000 | } | |
6001 | ||
a9d5adee JG |
6002 | struct uncharge_gather { |
6003 | struct mem_cgroup *memcg; | |
6004 | unsigned long pgpgout; | |
6005 | unsigned long nr_anon; | |
6006 | unsigned long nr_file; | |
6007 | unsigned long nr_kmem; | |
6008 | unsigned long nr_huge; | |
6009 | unsigned long nr_shmem; | |
6010 | struct page *dummy_page; | |
6011 | }; | |
6012 | ||
6013 | static inline void uncharge_gather_clear(struct uncharge_gather *ug) | |
747db954 | 6014 | { |
a9d5adee JG |
6015 | memset(ug, 0, sizeof(*ug)); |
6016 | } | |
6017 | ||
6018 | static void uncharge_batch(const struct uncharge_gather *ug) | |
6019 | { | |
6020 | unsigned long nr_pages = ug->nr_anon + ug->nr_file + ug->nr_kmem; | |
747db954 JW |
6021 | unsigned long flags; |
6022 | ||
a9d5adee JG |
6023 | if (!mem_cgroup_is_root(ug->memcg)) { |
6024 | page_counter_uncharge(&ug->memcg->memory, nr_pages); | |
7941d214 | 6025 | if (do_memsw_account()) |
a9d5adee JG |
6026 | page_counter_uncharge(&ug->memcg->memsw, nr_pages); |
6027 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && ug->nr_kmem) | |
6028 | page_counter_uncharge(&ug->memcg->kmem, ug->nr_kmem); | |
6029 | memcg_oom_recover(ug->memcg); | |
ce00a967 | 6030 | } |
747db954 JW |
6031 | |
6032 | local_irq_save(flags); | |
c9019e9b JW |
6033 | __mod_memcg_state(ug->memcg, MEMCG_RSS, -ug->nr_anon); |
6034 | __mod_memcg_state(ug->memcg, MEMCG_CACHE, -ug->nr_file); | |
6035 | __mod_memcg_state(ug->memcg, MEMCG_RSS_HUGE, -ug->nr_huge); | |
6036 | __mod_memcg_state(ug->memcg, NR_SHMEM, -ug->nr_shmem); | |
6037 | __count_memcg_events(ug->memcg, PGPGOUT, ug->pgpgout); | |
a983b5eb | 6038 | __this_cpu_add(ug->memcg->stat_cpu->nr_page_events, nr_pages); |
a9d5adee | 6039 | memcg_check_events(ug->memcg, ug->dummy_page); |
747db954 | 6040 | local_irq_restore(flags); |
e8ea14cc | 6041 | |
a9d5adee JG |
6042 | if (!mem_cgroup_is_root(ug->memcg)) |
6043 | css_put_many(&ug->memcg->css, nr_pages); | |
6044 | } | |
6045 | ||
6046 | static void uncharge_page(struct page *page, struct uncharge_gather *ug) | |
6047 | { | |
6048 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
3f2eb028 JG |
6049 | VM_BUG_ON_PAGE(page_count(page) && !is_zone_device_page(page) && |
6050 | !PageHWPoison(page) , page); | |
a9d5adee JG |
6051 | |
6052 | if (!page->mem_cgroup) | |
6053 | return; | |
6054 | ||
6055 | /* | |
6056 | * Nobody should be changing or seriously looking at | |
6057 | * page->mem_cgroup at this point, we have fully | |
6058 | * exclusive access to the page. | |
6059 | */ | |
6060 | ||
6061 | if (ug->memcg != page->mem_cgroup) { | |
6062 | if (ug->memcg) { | |
6063 | uncharge_batch(ug); | |
6064 | uncharge_gather_clear(ug); | |
6065 | } | |
6066 | ug->memcg = page->mem_cgroup; | |
6067 | } | |
6068 | ||
6069 | if (!PageKmemcg(page)) { | |
6070 | unsigned int nr_pages = 1; | |
6071 | ||
6072 | if (PageTransHuge(page)) { | |
6073 | nr_pages <<= compound_order(page); | |
6074 | ug->nr_huge += nr_pages; | |
6075 | } | |
6076 | if (PageAnon(page)) | |
6077 | ug->nr_anon += nr_pages; | |
6078 | else { | |
6079 | ug->nr_file += nr_pages; | |
6080 | if (PageSwapBacked(page)) | |
6081 | ug->nr_shmem += nr_pages; | |
6082 | } | |
6083 | ug->pgpgout++; | |
6084 | } else { | |
6085 | ug->nr_kmem += 1 << compound_order(page); | |
6086 | __ClearPageKmemcg(page); | |
6087 | } | |
6088 | ||
6089 | ug->dummy_page = page; | |
6090 | page->mem_cgroup = NULL; | |
747db954 JW |
6091 | } |
6092 | ||
6093 | static void uncharge_list(struct list_head *page_list) | |
6094 | { | |
a9d5adee | 6095 | struct uncharge_gather ug; |
747db954 | 6096 | struct list_head *next; |
a9d5adee JG |
6097 | |
6098 | uncharge_gather_clear(&ug); | |
747db954 | 6099 | |
8b592656 JW |
6100 | /* |
6101 | * Note that the list can be a single page->lru; hence the | |
6102 | * do-while loop instead of a simple list_for_each_entry(). | |
6103 | */ | |
747db954 JW |
6104 | next = page_list->next; |
6105 | do { | |
a9d5adee JG |
6106 | struct page *page; |
6107 | ||
747db954 JW |
6108 | page = list_entry(next, struct page, lru); |
6109 | next = page->lru.next; | |
6110 | ||
a9d5adee | 6111 | uncharge_page(page, &ug); |
747db954 JW |
6112 | } while (next != page_list); |
6113 | ||
a9d5adee JG |
6114 | if (ug.memcg) |
6115 | uncharge_batch(&ug); | |
747db954 JW |
6116 | } |
6117 | ||
0a31bc97 JW |
6118 | /** |
6119 | * mem_cgroup_uncharge - uncharge a page | |
6120 | * @page: page to uncharge | |
6121 | * | |
6122 | * Uncharge a page previously charged with mem_cgroup_try_charge() and | |
6123 | * mem_cgroup_commit_charge(). | |
6124 | */ | |
6125 | void mem_cgroup_uncharge(struct page *page) | |
6126 | { | |
a9d5adee JG |
6127 | struct uncharge_gather ug; |
6128 | ||
0a31bc97 JW |
6129 | if (mem_cgroup_disabled()) |
6130 | return; | |
6131 | ||
747db954 | 6132 | /* Don't touch page->lru of any random page, pre-check: */ |
1306a85a | 6133 | if (!page->mem_cgroup) |
0a31bc97 JW |
6134 | return; |
6135 | ||
a9d5adee JG |
6136 | uncharge_gather_clear(&ug); |
6137 | uncharge_page(page, &ug); | |
6138 | uncharge_batch(&ug); | |
747db954 | 6139 | } |
0a31bc97 | 6140 | |
747db954 JW |
6141 | /** |
6142 | * mem_cgroup_uncharge_list - uncharge a list of page | |
6143 | * @page_list: list of pages to uncharge | |
6144 | * | |
6145 | * Uncharge a list of pages previously charged with | |
6146 | * mem_cgroup_try_charge() and mem_cgroup_commit_charge(). | |
6147 | */ | |
6148 | void mem_cgroup_uncharge_list(struct list_head *page_list) | |
6149 | { | |
6150 | if (mem_cgroup_disabled()) | |
6151 | return; | |
0a31bc97 | 6152 | |
747db954 JW |
6153 | if (!list_empty(page_list)) |
6154 | uncharge_list(page_list); | |
0a31bc97 JW |
6155 | } |
6156 | ||
6157 | /** | |
6a93ca8f JW |
6158 | * mem_cgroup_migrate - charge a page's replacement |
6159 | * @oldpage: currently circulating page | |
6160 | * @newpage: replacement page | |
0a31bc97 | 6161 | * |
6a93ca8f JW |
6162 | * Charge @newpage as a replacement page for @oldpage. @oldpage will |
6163 | * be uncharged upon free. | |
0a31bc97 JW |
6164 | * |
6165 | * Both pages must be locked, @newpage->mapping must be set up. | |
6166 | */ | |
6a93ca8f | 6167 | void mem_cgroup_migrate(struct page *oldpage, struct page *newpage) |
0a31bc97 | 6168 | { |
29833315 | 6169 | struct mem_cgroup *memcg; |
44b7a8d3 JW |
6170 | unsigned int nr_pages; |
6171 | bool compound; | |
d93c4130 | 6172 | unsigned long flags; |
0a31bc97 JW |
6173 | |
6174 | VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage); | |
6175 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); | |
0a31bc97 | 6176 | VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage); |
6abb5a86 JW |
6177 | VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage), |
6178 | newpage); | |
0a31bc97 JW |
6179 | |
6180 | if (mem_cgroup_disabled()) | |
6181 | return; | |
6182 | ||
6183 | /* Page cache replacement: new page already charged? */ | |
1306a85a | 6184 | if (newpage->mem_cgroup) |
0a31bc97 JW |
6185 | return; |
6186 | ||
45637bab | 6187 | /* Swapcache readahead pages can get replaced before being charged */ |
1306a85a | 6188 | memcg = oldpage->mem_cgroup; |
29833315 | 6189 | if (!memcg) |
0a31bc97 JW |
6190 | return; |
6191 | ||
44b7a8d3 JW |
6192 | /* Force-charge the new page. The old one will be freed soon */ |
6193 | compound = PageTransHuge(newpage); | |
6194 | nr_pages = compound ? hpage_nr_pages(newpage) : 1; | |
6195 | ||
6196 | page_counter_charge(&memcg->memory, nr_pages); | |
6197 | if (do_memsw_account()) | |
6198 | page_counter_charge(&memcg->memsw, nr_pages); | |
6199 | css_get_many(&memcg->css, nr_pages); | |
0a31bc97 | 6200 | |
9cf7666a | 6201 | commit_charge(newpage, memcg, false); |
44b7a8d3 | 6202 | |
d93c4130 | 6203 | local_irq_save(flags); |
44b7a8d3 JW |
6204 | mem_cgroup_charge_statistics(memcg, newpage, compound, nr_pages); |
6205 | memcg_check_events(memcg, newpage); | |
d93c4130 | 6206 | local_irq_restore(flags); |
0a31bc97 JW |
6207 | } |
6208 | ||
ef12947c | 6209 | DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key); |
11092087 JW |
6210 | EXPORT_SYMBOL(memcg_sockets_enabled_key); |
6211 | ||
2d758073 | 6212 | void mem_cgroup_sk_alloc(struct sock *sk) |
11092087 JW |
6213 | { |
6214 | struct mem_cgroup *memcg; | |
6215 | ||
2d758073 JW |
6216 | if (!mem_cgroup_sockets_enabled) |
6217 | return; | |
6218 | ||
edbe69ef RG |
6219 | /* |
6220 | * Socket cloning can throw us here with sk_memcg already | |
6221 | * filled. It won't however, necessarily happen from | |
6222 | * process context. So the test for root memcg given | |
6223 | * the current task's memcg won't help us in this case. | |
6224 | * | |
6225 | * Respecting the original socket's memcg is a better | |
6226 | * decision in this case. | |
6227 | */ | |
6228 | if (sk->sk_memcg) { | |
6229 | css_get(&sk->sk_memcg->css); | |
6230 | return; | |
6231 | } | |
6232 | ||
11092087 JW |
6233 | rcu_read_lock(); |
6234 | memcg = mem_cgroup_from_task(current); | |
f7e1cb6e JW |
6235 | if (memcg == root_mem_cgroup) |
6236 | goto out; | |
0db15298 | 6237 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active) |
f7e1cb6e | 6238 | goto out; |
f7e1cb6e | 6239 | if (css_tryget_online(&memcg->css)) |
11092087 | 6240 | sk->sk_memcg = memcg; |
f7e1cb6e | 6241 | out: |
11092087 JW |
6242 | rcu_read_unlock(); |
6243 | } | |
11092087 | 6244 | |
2d758073 | 6245 | void mem_cgroup_sk_free(struct sock *sk) |
11092087 | 6246 | { |
2d758073 JW |
6247 | if (sk->sk_memcg) |
6248 | css_put(&sk->sk_memcg->css); | |
11092087 JW |
6249 | } |
6250 | ||
6251 | /** | |
6252 | * mem_cgroup_charge_skmem - charge socket memory | |
6253 | * @memcg: memcg to charge | |
6254 | * @nr_pages: number of pages to charge | |
6255 | * | |
6256 | * Charges @nr_pages to @memcg. Returns %true if the charge fit within | |
6257 | * @memcg's configured limit, %false if the charge had to be forced. | |
6258 | */ | |
6259 | bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
6260 | { | |
f7e1cb6e | 6261 | gfp_t gfp_mask = GFP_KERNEL; |
11092087 | 6262 | |
f7e1cb6e | 6263 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 6264 | struct page_counter *fail; |
f7e1cb6e | 6265 | |
0db15298 JW |
6266 | if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) { |
6267 | memcg->tcpmem_pressure = 0; | |
f7e1cb6e JW |
6268 | return true; |
6269 | } | |
0db15298 JW |
6270 | page_counter_charge(&memcg->tcpmem, nr_pages); |
6271 | memcg->tcpmem_pressure = 1; | |
f7e1cb6e | 6272 | return false; |
11092087 | 6273 | } |
d886f4e4 | 6274 | |
f7e1cb6e JW |
6275 | /* Don't block in the packet receive path */ |
6276 | if (in_softirq()) | |
6277 | gfp_mask = GFP_NOWAIT; | |
6278 | ||
c9019e9b | 6279 | mod_memcg_state(memcg, MEMCG_SOCK, nr_pages); |
b2807f07 | 6280 | |
f7e1cb6e JW |
6281 | if (try_charge(memcg, gfp_mask, nr_pages) == 0) |
6282 | return true; | |
6283 | ||
6284 | try_charge(memcg, gfp_mask|__GFP_NOFAIL, nr_pages); | |
11092087 JW |
6285 | return false; |
6286 | } | |
6287 | ||
6288 | /** | |
6289 | * mem_cgroup_uncharge_skmem - uncharge socket memory | |
b7701a5f MR |
6290 | * @memcg: memcg to uncharge |
6291 | * @nr_pages: number of pages to uncharge | |
11092087 JW |
6292 | */ |
6293 | void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
6294 | { | |
f7e1cb6e | 6295 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 6296 | page_counter_uncharge(&memcg->tcpmem, nr_pages); |
f7e1cb6e JW |
6297 | return; |
6298 | } | |
d886f4e4 | 6299 | |
c9019e9b | 6300 | mod_memcg_state(memcg, MEMCG_SOCK, -nr_pages); |
b2807f07 | 6301 | |
475d0487 | 6302 | refill_stock(memcg, nr_pages); |
11092087 JW |
6303 | } |
6304 | ||
f7e1cb6e JW |
6305 | static int __init cgroup_memory(char *s) |
6306 | { | |
6307 | char *token; | |
6308 | ||
6309 | while ((token = strsep(&s, ",")) != NULL) { | |
6310 | if (!*token) | |
6311 | continue; | |
6312 | if (!strcmp(token, "nosocket")) | |
6313 | cgroup_memory_nosocket = true; | |
04823c83 VD |
6314 | if (!strcmp(token, "nokmem")) |
6315 | cgroup_memory_nokmem = true; | |
f7e1cb6e JW |
6316 | } |
6317 | return 0; | |
6318 | } | |
6319 | __setup("cgroup.memory=", cgroup_memory); | |
11092087 | 6320 | |
2d11085e | 6321 | /* |
1081312f MH |
6322 | * subsys_initcall() for memory controller. |
6323 | * | |
308167fc SAS |
6324 | * Some parts like memcg_hotplug_cpu_dead() have to be initialized from this |
6325 | * context because of lock dependencies (cgroup_lock -> cpu hotplug) but | |
6326 | * basically everything that doesn't depend on a specific mem_cgroup structure | |
6327 | * should be initialized from here. | |
2d11085e MH |
6328 | */ |
6329 | static int __init mem_cgroup_init(void) | |
6330 | { | |
95a045f6 JW |
6331 | int cpu, node; |
6332 | ||
84c07d11 | 6333 | #ifdef CONFIG_MEMCG_KMEM |
13583c3d VD |
6334 | /* |
6335 | * Kmem cache creation is mostly done with the slab_mutex held, | |
17cc4dfe TH |
6336 | * so use a workqueue with limited concurrency to avoid stalling |
6337 | * all worker threads in case lots of cgroups are created and | |
6338 | * destroyed simultaneously. | |
13583c3d | 6339 | */ |
17cc4dfe TH |
6340 | memcg_kmem_cache_wq = alloc_workqueue("memcg_kmem_cache", 0, 1); |
6341 | BUG_ON(!memcg_kmem_cache_wq); | |
13583c3d VD |
6342 | #endif |
6343 | ||
308167fc SAS |
6344 | cpuhp_setup_state_nocalls(CPUHP_MM_MEMCQ_DEAD, "mm/memctrl:dead", NULL, |
6345 | memcg_hotplug_cpu_dead); | |
95a045f6 JW |
6346 | |
6347 | for_each_possible_cpu(cpu) | |
6348 | INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work, | |
6349 | drain_local_stock); | |
6350 | ||
6351 | for_each_node(node) { | |
6352 | struct mem_cgroup_tree_per_node *rtpn; | |
95a045f6 JW |
6353 | |
6354 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, | |
6355 | node_online(node) ? node : NUMA_NO_NODE); | |
6356 | ||
ef8f2327 | 6357 | rtpn->rb_root = RB_ROOT; |
fa90b2fd | 6358 | rtpn->rb_rightmost = NULL; |
ef8f2327 | 6359 | spin_lock_init(&rtpn->lock); |
95a045f6 JW |
6360 | soft_limit_tree.rb_tree_per_node[node] = rtpn; |
6361 | } | |
6362 | ||
2d11085e MH |
6363 | return 0; |
6364 | } | |
6365 | subsys_initcall(mem_cgroup_init); | |
21afa38e JW |
6366 | |
6367 | #ifdef CONFIG_MEMCG_SWAP | |
358c07fc AB |
6368 | static struct mem_cgroup *mem_cgroup_id_get_online(struct mem_cgroup *memcg) |
6369 | { | |
1c2d479a | 6370 | while (!refcount_inc_not_zero(&memcg->id.ref)) { |
358c07fc AB |
6371 | /* |
6372 | * The root cgroup cannot be destroyed, so it's refcount must | |
6373 | * always be >= 1. | |
6374 | */ | |
6375 | if (WARN_ON_ONCE(memcg == root_mem_cgroup)) { | |
6376 | VM_BUG_ON(1); | |
6377 | break; | |
6378 | } | |
6379 | memcg = parent_mem_cgroup(memcg); | |
6380 | if (!memcg) | |
6381 | memcg = root_mem_cgroup; | |
6382 | } | |
6383 | return memcg; | |
6384 | } | |
6385 | ||
21afa38e JW |
6386 | /** |
6387 | * mem_cgroup_swapout - transfer a memsw charge to swap | |
6388 | * @page: page whose memsw charge to transfer | |
6389 | * @entry: swap entry to move the charge to | |
6390 | * | |
6391 | * Transfer the memsw charge of @page to @entry. | |
6392 | */ | |
6393 | void mem_cgroup_swapout(struct page *page, swp_entry_t entry) | |
6394 | { | |
1f47b61f | 6395 | struct mem_cgroup *memcg, *swap_memcg; |
d6810d73 | 6396 | unsigned int nr_entries; |
21afa38e JW |
6397 | unsigned short oldid; |
6398 | ||
6399 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
6400 | VM_BUG_ON_PAGE(page_count(page), page); | |
6401 | ||
7941d214 | 6402 | if (!do_memsw_account()) |
21afa38e JW |
6403 | return; |
6404 | ||
6405 | memcg = page->mem_cgroup; | |
6406 | ||
6407 | /* Readahead page, never charged */ | |
6408 | if (!memcg) | |
6409 | return; | |
6410 | ||
1f47b61f VD |
6411 | /* |
6412 | * In case the memcg owning these pages has been offlined and doesn't | |
6413 | * have an ID allocated to it anymore, charge the closest online | |
6414 | * ancestor for the swap instead and transfer the memory+swap charge. | |
6415 | */ | |
6416 | swap_memcg = mem_cgroup_id_get_online(memcg); | |
d6810d73 HY |
6417 | nr_entries = hpage_nr_pages(page); |
6418 | /* Get references for the tail pages, too */ | |
6419 | if (nr_entries > 1) | |
6420 | mem_cgroup_id_get_many(swap_memcg, nr_entries - 1); | |
6421 | oldid = swap_cgroup_record(entry, mem_cgroup_id(swap_memcg), | |
6422 | nr_entries); | |
21afa38e | 6423 | VM_BUG_ON_PAGE(oldid, page); |
c9019e9b | 6424 | mod_memcg_state(swap_memcg, MEMCG_SWAP, nr_entries); |
21afa38e JW |
6425 | |
6426 | page->mem_cgroup = NULL; | |
6427 | ||
6428 | if (!mem_cgroup_is_root(memcg)) | |
d6810d73 | 6429 | page_counter_uncharge(&memcg->memory, nr_entries); |
21afa38e | 6430 | |
1f47b61f VD |
6431 | if (memcg != swap_memcg) { |
6432 | if (!mem_cgroup_is_root(swap_memcg)) | |
d6810d73 HY |
6433 | page_counter_charge(&swap_memcg->memsw, nr_entries); |
6434 | page_counter_uncharge(&memcg->memsw, nr_entries); | |
1f47b61f VD |
6435 | } |
6436 | ||
ce9ce665 SAS |
6437 | /* |
6438 | * Interrupts should be disabled here because the caller holds the | |
b93b0163 | 6439 | * i_pages lock which is taken with interrupts-off. It is |
ce9ce665 | 6440 | * important here to have the interrupts disabled because it is the |
b93b0163 | 6441 | * only synchronisation we have for updating the per-CPU variables. |
ce9ce665 SAS |
6442 | */ |
6443 | VM_BUG_ON(!irqs_disabled()); | |
d6810d73 HY |
6444 | mem_cgroup_charge_statistics(memcg, page, PageTransHuge(page), |
6445 | -nr_entries); | |
21afa38e | 6446 | memcg_check_events(memcg, page); |
73f576c0 JW |
6447 | |
6448 | if (!mem_cgroup_is_root(memcg)) | |
d08afa14 | 6449 | css_put_many(&memcg->css, nr_entries); |
21afa38e JW |
6450 | } |
6451 | ||
38d8b4e6 HY |
6452 | /** |
6453 | * mem_cgroup_try_charge_swap - try charging swap space for a page | |
37e84351 VD |
6454 | * @page: page being added to swap |
6455 | * @entry: swap entry to charge | |
6456 | * | |
38d8b4e6 | 6457 | * Try to charge @page's memcg for the swap space at @entry. |
37e84351 VD |
6458 | * |
6459 | * Returns 0 on success, -ENOMEM on failure. | |
6460 | */ | |
6461 | int mem_cgroup_try_charge_swap(struct page *page, swp_entry_t entry) | |
6462 | { | |
38d8b4e6 | 6463 | unsigned int nr_pages = hpage_nr_pages(page); |
37e84351 | 6464 | struct page_counter *counter; |
38d8b4e6 | 6465 | struct mem_cgroup *memcg; |
37e84351 VD |
6466 | unsigned short oldid; |
6467 | ||
6468 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) || !do_swap_account) | |
6469 | return 0; | |
6470 | ||
6471 | memcg = page->mem_cgroup; | |
6472 | ||
6473 | /* Readahead page, never charged */ | |
6474 | if (!memcg) | |
6475 | return 0; | |
6476 | ||
f3a53a3a TH |
6477 | if (!entry.val) { |
6478 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); | |
bb98f2c5 | 6479 | return 0; |
f3a53a3a | 6480 | } |
bb98f2c5 | 6481 | |
1f47b61f VD |
6482 | memcg = mem_cgroup_id_get_online(memcg); |
6483 | ||
37e84351 | 6484 | if (!mem_cgroup_is_root(memcg) && |
38d8b4e6 | 6485 | !page_counter_try_charge(&memcg->swap, nr_pages, &counter)) { |
f3a53a3a TH |
6486 | memcg_memory_event(memcg, MEMCG_SWAP_MAX); |
6487 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); | |
1f47b61f | 6488 | mem_cgroup_id_put(memcg); |
37e84351 | 6489 | return -ENOMEM; |
1f47b61f | 6490 | } |
37e84351 | 6491 | |
38d8b4e6 HY |
6492 | /* Get references for the tail pages, too */ |
6493 | if (nr_pages > 1) | |
6494 | mem_cgroup_id_get_many(memcg, nr_pages - 1); | |
6495 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg), nr_pages); | |
37e84351 | 6496 | VM_BUG_ON_PAGE(oldid, page); |
c9019e9b | 6497 | mod_memcg_state(memcg, MEMCG_SWAP, nr_pages); |
37e84351 | 6498 | |
37e84351 VD |
6499 | return 0; |
6500 | } | |
6501 | ||
21afa38e | 6502 | /** |
38d8b4e6 | 6503 | * mem_cgroup_uncharge_swap - uncharge swap space |
21afa38e | 6504 | * @entry: swap entry to uncharge |
38d8b4e6 | 6505 | * @nr_pages: the amount of swap space to uncharge |
21afa38e | 6506 | */ |
38d8b4e6 | 6507 | void mem_cgroup_uncharge_swap(swp_entry_t entry, unsigned int nr_pages) |
21afa38e JW |
6508 | { |
6509 | struct mem_cgroup *memcg; | |
6510 | unsigned short id; | |
6511 | ||
37e84351 | 6512 | if (!do_swap_account) |
21afa38e JW |
6513 | return; |
6514 | ||
38d8b4e6 | 6515 | id = swap_cgroup_record(entry, 0, nr_pages); |
21afa38e | 6516 | rcu_read_lock(); |
adbe427b | 6517 | memcg = mem_cgroup_from_id(id); |
21afa38e | 6518 | if (memcg) { |
37e84351 VD |
6519 | if (!mem_cgroup_is_root(memcg)) { |
6520 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
38d8b4e6 | 6521 | page_counter_uncharge(&memcg->swap, nr_pages); |
37e84351 | 6522 | else |
38d8b4e6 | 6523 | page_counter_uncharge(&memcg->memsw, nr_pages); |
37e84351 | 6524 | } |
c9019e9b | 6525 | mod_memcg_state(memcg, MEMCG_SWAP, -nr_pages); |
38d8b4e6 | 6526 | mem_cgroup_id_put_many(memcg, nr_pages); |
21afa38e JW |
6527 | } |
6528 | rcu_read_unlock(); | |
6529 | } | |
6530 | ||
d8b38438 VD |
6531 | long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg) |
6532 | { | |
6533 | long nr_swap_pages = get_nr_swap_pages(); | |
6534 | ||
6535 | if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
6536 | return nr_swap_pages; | |
6537 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) | |
6538 | nr_swap_pages = min_t(long, nr_swap_pages, | |
bbec2e15 | 6539 | READ_ONCE(memcg->swap.max) - |
d8b38438 VD |
6540 | page_counter_read(&memcg->swap)); |
6541 | return nr_swap_pages; | |
6542 | } | |
6543 | ||
5ccc5aba VD |
6544 | bool mem_cgroup_swap_full(struct page *page) |
6545 | { | |
6546 | struct mem_cgroup *memcg; | |
6547 | ||
6548 | VM_BUG_ON_PAGE(!PageLocked(page), page); | |
6549 | ||
6550 | if (vm_swap_full()) | |
6551 | return true; | |
6552 | if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
6553 | return false; | |
6554 | ||
6555 | memcg = page->mem_cgroup; | |
6556 | if (!memcg) | |
6557 | return false; | |
6558 | ||
6559 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) | |
bbec2e15 | 6560 | if (page_counter_read(&memcg->swap) * 2 >= memcg->swap.max) |
5ccc5aba VD |
6561 | return true; |
6562 | ||
6563 | return false; | |
6564 | } | |
6565 | ||
21afa38e JW |
6566 | /* for remember boot option*/ |
6567 | #ifdef CONFIG_MEMCG_SWAP_ENABLED | |
6568 | static int really_do_swap_account __initdata = 1; | |
6569 | #else | |
6570 | static int really_do_swap_account __initdata; | |
6571 | #endif | |
6572 | ||
6573 | static int __init enable_swap_account(char *s) | |
6574 | { | |
6575 | if (!strcmp(s, "1")) | |
6576 | really_do_swap_account = 1; | |
6577 | else if (!strcmp(s, "0")) | |
6578 | really_do_swap_account = 0; | |
6579 | return 1; | |
6580 | } | |
6581 | __setup("swapaccount=", enable_swap_account); | |
6582 | ||
37e84351 VD |
6583 | static u64 swap_current_read(struct cgroup_subsys_state *css, |
6584 | struct cftype *cft) | |
6585 | { | |
6586 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
6587 | ||
6588 | return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE; | |
6589 | } | |
6590 | ||
6591 | static int swap_max_show(struct seq_file *m, void *v) | |
6592 | { | |
677dc973 CD |
6593 | return seq_puts_memcg_tunable(m, |
6594 | READ_ONCE(mem_cgroup_from_seq(m)->swap.max)); | |
37e84351 VD |
6595 | } |
6596 | ||
6597 | static ssize_t swap_max_write(struct kernfs_open_file *of, | |
6598 | char *buf, size_t nbytes, loff_t off) | |
6599 | { | |
6600 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6601 | unsigned long max; | |
6602 | int err; | |
6603 | ||
6604 | buf = strstrip(buf); | |
6605 | err = page_counter_memparse(buf, "max", &max); | |
6606 | if (err) | |
6607 | return err; | |
6608 | ||
be09102b | 6609 | xchg(&memcg->swap.max, max); |
37e84351 VD |
6610 | |
6611 | return nbytes; | |
6612 | } | |
6613 | ||
f3a53a3a TH |
6614 | static int swap_events_show(struct seq_file *m, void *v) |
6615 | { | |
aa9694bb | 6616 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
f3a53a3a TH |
6617 | |
6618 | seq_printf(m, "max %lu\n", | |
6619 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_MAX])); | |
6620 | seq_printf(m, "fail %lu\n", | |
6621 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_FAIL])); | |
6622 | ||
6623 | return 0; | |
6624 | } | |
6625 | ||
37e84351 VD |
6626 | static struct cftype swap_files[] = { |
6627 | { | |
6628 | .name = "swap.current", | |
6629 | .flags = CFTYPE_NOT_ON_ROOT, | |
6630 | .read_u64 = swap_current_read, | |
6631 | }, | |
6632 | { | |
6633 | .name = "swap.max", | |
6634 | .flags = CFTYPE_NOT_ON_ROOT, | |
6635 | .seq_show = swap_max_show, | |
6636 | .write = swap_max_write, | |
6637 | }, | |
f3a53a3a TH |
6638 | { |
6639 | .name = "swap.events", | |
6640 | .flags = CFTYPE_NOT_ON_ROOT, | |
6641 | .file_offset = offsetof(struct mem_cgroup, swap_events_file), | |
6642 | .seq_show = swap_events_show, | |
6643 | }, | |
37e84351 VD |
6644 | { } /* terminate */ |
6645 | }; | |
6646 | ||
21afa38e JW |
6647 | static struct cftype memsw_cgroup_files[] = { |
6648 | { | |
6649 | .name = "memsw.usage_in_bytes", | |
6650 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
6651 | .read_u64 = mem_cgroup_read_u64, | |
6652 | }, | |
6653 | { | |
6654 | .name = "memsw.max_usage_in_bytes", | |
6655 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
6656 | .write = mem_cgroup_reset, | |
6657 | .read_u64 = mem_cgroup_read_u64, | |
6658 | }, | |
6659 | { | |
6660 | .name = "memsw.limit_in_bytes", | |
6661 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
6662 | .write = mem_cgroup_write, | |
6663 | .read_u64 = mem_cgroup_read_u64, | |
6664 | }, | |
6665 | { | |
6666 | .name = "memsw.failcnt", | |
6667 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
6668 | .write = mem_cgroup_reset, | |
6669 | .read_u64 = mem_cgroup_read_u64, | |
6670 | }, | |
6671 | { }, /* terminate */ | |
6672 | }; | |
6673 | ||
6674 | static int __init mem_cgroup_swap_init(void) | |
6675 | { | |
6676 | if (!mem_cgroup_disabled() && really_do_swap_account) { | |
6677 | do_swap_account = 1; | |
37e84351 VD |
6678 | WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys, |
6679 | swap_files)); | |
21afa38e JW |
6680 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, |
6681 | memsw_cgroup_files)); | |
6682 | } | |
6683 | return 0; | |
6684 | } | |
6685 | subsys_initcall(mem_cgroup_swap_init); | |
6686 | ||
6687 | #endif /* CONFIG_MEMCG_SWAP */ |